Compare commits

..

5 Commits

Author SHA1 Message Date
770761b3f9 docs: update vide docs 2025-08-04 19:30:48 -06:00
f191811e23 fix: daily.co initial support works 2025-08-04 19:06:15 -06:00
6b3c193672 docs: update vibe docs 2025-08-04 18:50:55 -06:00
06869ef5ca fix: alembic upgrade 2025-08-04 11:15:43 -06:00
8b644384a2 chore: remove refactor md (#527) 2025-08-01 18:22:50 -06:00
406 changed files with 19969 additions and 59896 deletions

View File

@@ -2,8 +2,6 @@ name: Test Database Migrations
on:
push:
branches:
- main
paths:
- "server/migrations/**"
- "server/reflector/db/**"
@@ -19,43 +17,10 @@ on:
jobs:
test-migrations:
runs-on: ubuntu-latest
concurrency:
group: db-ubuntu-latest-${{ github.ref }}
cancel-in-progress: true
services:
postgres:
image: postgres:17
env:
POSTGRES_USER: reflector
POSTGRES_PASSWORD: reflector
POSTGRES_DB: reflector
ports:
- 5432:5432
options: >-
--health-cmd pg_isready -h 127.0.0.1 -p 5432
--health-interval 10s
--health-timeout 5s
--health-retries 5
env:
DATABASE_URL: postgresql://reflector:reflector@localhost:5432/reflector
steps:
- uses: actions/checkout@v4
- name: Install PostgreSQL client
run: sudo apt-get update && sudo apt-get install -y postgresql-client | cat
- name: Wait for Postgres
run: |
for i in {1..30}; do
if pg_isready -h localhost -p 5432; then
echo "Postgres is ready"
break
fi
echo "Waiting for Postgres... ($i)" && sleep 1
done
- name: Install uv
uses: astral-sh/setup-uv@v3
with:

47
.github/workflows/deploy.yml vendored Normal file
View File

@@ -0,0 +1,47 @@
name: Deploy to Amazon ECS
on: [workflow_dispatch]
env:
# 950402358378.dkr.ecr.us-east-1.amazonaws.com/reflector
AWS_REGION: us-east-1
ECR_REPOSITORY: reflector
jobs:
deploy:
runs-on: ubuntu-latest
permissions:
deployments: write
contents: read
steps:
- uses: actions/checkout@v3
- name: Configure AWS credentials
uses: aws-actions/configure-aws-credentials@0e613a0980cbf65ed5b322eb7a1e075d28913a83
with:
aws-access-key-id: ${{ secrets.AWS_ACCESS_KEY_ID }}
aws-secret-access-key: ${{ secrets.AWS_SECRET_ACCESS_KEY }}
aws-region: ${{ env.AWS_REGION }}
- name: Login to Amazon ECR
id: login-ecr
uses: aws-actions/amazon-ecr-login@62f4f872db3836360b72999f4b87f1ff13310f3a
- name: Set up QEMU
uses: docker/setup-qemu-action@v2
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v2
- name: Build and push
id: docker_build
uses: docker/build-push-action@v4
with:
context: server
platforms: linux/amd64,linux/arm64
push: true
tags: ${{ steps.login-ecr.outputs.registry }}/${{ env.ECR_REPOSITORY }}:latest
cache-from: type=gha
cache-to: type=gha,mode=max

View File

@@ -1,53 +0,0 @@
name: Build and Push Backend Docker Image (Docker Hub)
on:
push:
tags:
- "v*"
workflow_dispatch:
env:
REGISTRY: docker.io
IMAGE_NAME: monadicalsas/reflector-backend
jobs:
build-and-push:
runs-on: ubuntu-latest
permissions:
contents: read
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Log in to Docker Hub
uses: docker/login-action@v3
with:
registry: ${{ env.REGISTRY }}
username: monadicalsas
password: ${{ secrets.DOCKERHUB_TOKEN }}
- name: Extract metadata
id: meta
uses: docker/metadata-action@v5
with:
images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
tags: |
type=ref,event=branch
type=ref,event=tag
type=raw,value=latest,enable={{is_default_branch}}
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v3
- name: Build and push Docker image
uses: docker/build-push-action@v5
with:
context: ./server
file: ./server/Dockerfile
push: true
tags: ${{ steps.meta.outputs.tags }}
labels: ${{ steps.meta.outputs.labels }}
cache-from: type=gha
cache-to: type=gha,mode=max
platforms: linux/amd64,linux/arm64

View File

@@ -1,70 +0,0 @@
name: Build and Push Frontend Docker Image
on:
push:
tags:
- "v*"
workflow_dispatch:
env:
REGISTRY: docker.io
IMAGE_NAME: monadicalsas/reflector-frontend
jobs:
build-and-push:
runs-on: ubuntu-latest
permissions:
contents: read
steps:
- name: Checkout repository
uses: actions/checkout@v4
- name: Log in to Docker Hub
uses: docker/login-action@v3
with:
registry: ${{ env.REGISTRY }}
username: monadicalsas
password: ${{ secrets.DOCKERHUB_TOKEN }}
- name: Extract metadata
id: meta
uses: docker/metadata-action@v5
with:
images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
tags: |
type=ref,event=branch
type=ref,event=tag
type=raw,value=latest,enable={{is_default_branch}}
github-token: ${{ secrets.GITHUB_TOKEN }}
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v3
- name: Build and push Docker image
uses: docker/build-push-action@v5
with:
context: ./www
file: ./www/Dockerfile
push: true
tags: ${{ steps.meta.outputs.tags }}
labels: ${{ steps.meta.outputs.labels }}
cache-from: type=gha
cache-to: type=gha,mode=max
platforms: linux/amd64,linux/arm64
deploy:
needs: build-and-push
runs-on: ubuntu-latest
if: success()
strategy:
matrix:
environment: [reflector-monadical, reflector-media]
environment: ${{ matrix.environment }}
steps:
- name: Trigger Coolify deployment
run: |
curl -X POST "${{ secrets.COOLIFY_WEBHOOK_URL }}" \
-H "Content-Type: application/json" \
-H "Authorization: Bearer ${{ secrets.COOLIFY_WEBHOOK_TOKEN }}" \
-f || (echo "Failed to trigger Coolify deployment for ${{ matrix.environment }}" && exit 1)

View File

@@ -1,24 +0,0 @@
name: pre-commit
on:
pull_request:
push:
branches: [main]
jobs:
pre-commit:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v5
- uses: actions/setup-python@v5
- uses: pnpm/action-setup@v4
with:
version: 10
- uses: actions/setup-node@v4
with:
node-version: 22
cache: "pnpm"
cache-dependency-path: "www/pnpm-lock.yaml"
- name: Install dependencies
run: cd www && pnpm install --frozen-lockfile
- uses: pre-commit/action@v3.0.1

View File

@@ -1,45 +0,0 @@
name: Test Next Server
on:
pull_request:
paths:
- "www/**"
push:
branches:
- main
paths:
- "www/**"
jobs:
test-next-server:
runs-on: ubuntu-latest
defaults:
run:
working-directory: ./www
steps:
- uses: actions/checkout@v4
- name: Setup Node.js
uses: actions/setup-node@v4
with:
node-version: '20'
- name: Install pnpm
uses: pnpm/action-setup@v4
with:
version: 8
- name: Setup Node.js cache
uses: actions/setup-node@v4
with:
node-version: '20'
cache: 'pnpm'
cache-dependency-path: './www/pnpm-lock.yaml'
- name: Install dependencies
run: pnpm install
- name: Run tests
run: pnpm test

View File

@@ -5,17 +5,12 @@ on:
paths:
- "server/**"
push:
branches:
- main
paths:
- "server/**"
jobs:
pytest:
runs-on: ubuntu-latest
concurrency:
group: pytest-${{ github.ref }}
cancel-in-progress: true
services:
redis:
image: redis:6
@@ -24,47 +19,29 @@ jobs:
steps:
- uses: actions/checkout@v4
- name: Install uv
uses: astral-sh/setup-uv@v6
uses: astral-sh/setup-uv@v3
with:
enable-cache: true
working-directory: server
- name: Tests
run: |
cd server
uv run -m pytest -v tests
docker-amd64:
runs-on: linux-amd64
concurrency:
group: docker-amd64-${{ github.ref }}
cancel-in-progress: true
docker:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v4
- name: Set up QEMU
uses: docker/setup-qemu-action@v2
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v3
- name: Build AMD64
uses: docker/build-push-action@v6
uses: docker/setup-buildx-action@v2
- name: Build and push
id: docker_build
uses: docker/build-push-action@v4
with:
context: server
platforms: linux/amd64
cache-from: type=gha,scope=amd64
cache-to: type=gha,mode=max,scope=amd64
github-token: ${{ secrets.GHA_CACHE_TOKEN }}
docker-arm64:
runs-on: linux-arm64
concurrency:
group: docker-arm64-${{ github.ref }}
cancel-in-progress: true
steps:
- uses: actions/checkout@v4
- name: Set up Docker Buildx
uses: docker/setup-buildx-action@v3
- name: Build ARM64
uses: docker/build-push-action@v6
with:
context: server
platforms: linux/arm64
cache-from: type=gha,scope=arm64
cache-to: type=gha,mode=max,scope=arm64
github-token: ${{ secrets.GHA_CACHE_TOKEN }}
platforms: linux/amd64,linux/arm64
cache-from: type=gha
cache-to: type=gha,mode=max

6
.gitignore vendored
View File

@@ -13,9 +13,3 @@ restart-dev.sh
data/
www/REFACTOR.md
www/reload-frontend
server/test.sqlite
CLAUDE.local.md
www/.env.development
www/.env.production
.playwright-mcp
.secrets

View File

@@ -1 +0,0 @@
b9d891d3424f371642cb032ecfd0e2564470a72c:server/tests/test_transcripts_recording_deletion.py:generic-api-key:15

View File

@@ -3,10 +3,10 @@
repos:
- repo: local
hooks:
- id: format
name: run format
- id: yarn-format
name: run yarn format
language: system
entry: bash -c 'cd www && pnpm format'
entry: bash -c 'cd www && yarn format'
pass_filenames: false
files: ^www/
@@ -23,12 +23,8 @@ repos:
- id: ruff
args:
- --fix
# Uses select rules from server/pyproject.toml
- --select
- I,F401
files: ^server/
- id: ruff-format
files: ^server/
- repo: https://github.com/gitleaks/gitleaks
rev: v8.28.0
hooks:
- id: gitleaks

View File

@@ -1,24 +0,0 @@
# Example secrets file for GitHub Actions workflows
# Copy this to .secrets and fill in your values
# These secrets should be configured in GitHub repository settings:
# Settings > Secrets and variables > Actions
# DockerHub Configuration (required for frontend and backend deployment)
# Create a Docker Hub access token at https://hub.docker.com/settings/security
# Username: monadicalsas
DOCKERHUB_TOKEN=your-dockerhub-access-token
# GitHub Token (required for frontend and backend deployment)
# Used by docker/metadata-action for extracting image metadata
# Can use the default GITHUB_TOKEN or create a personal access token
GITHUB_TOKEN=your-github-token-or-use-default-GITHUB_TOKEN
# Coolify Deployment Webhook (required for frontend deployment)
# Used to trigger automatic deployment after image push
# Configure these secrets in GitHub Environments:
# Each environment should have:
# - COOLIFY_WEBHOOK_URL: The webhook URL for that specific deployment
# - COOLIFY_WEBHOOK_TOKEN: The webhook token (can be the same for both if using same token)
# Optional: GitHub Actions Cache Token (for local testing with act)
GHA_CACHE_TOKEN=your-github-token-or-empty

View File

@@ -1,349 +1,5 @@
# Changelog
## [0.24.0](https://github.com/Monadical-SAS/reflector/compare/v0.23.2...v0.24.0) (2025-12-18)
### Features
* identify action items ([#790](https://github.com/Monadical-SAS/reflector/issues/790)) ([964cd78](https://github.com/Monadical-SAS/reflector/commit/964cd78bb699d83d012ae4b8c96565df25b90a5d))
### Bug Fixes
* automatically reprocess daily recordings ([#797](https://github.com/Monadical-SAS/reflector/issues/797)) ([5f458aa](https://github.com/Monadical-SAS/reflector/commit/5f458aa4a7ec3d00ca5ec49d62fcc8ad232b138e))
* daily video optimisation ([#789](https://github.com/Monadical-SAS/reflector/issues/789)) ([16284e1](https://github.com/Monadical-SAS/reflector/commit/16284e1ac3faede2b74f0d91b50c0b5612af2c35))
* main menu login ([#800](https://github.com/Monadical-SAS/reflector/issues/800)) ([0bc971b](https://github.com/Monadical-SAS/reflector/commit/0bc971ba966a52d719c8c240b47dc7b3bdea4391))
* retry on workflow timeout ([#798](https://github.com/Monadical-SAS/reflector/issues/798)) ([5f7dfad](https://github.com/Monadical-SAS/reflector/commit/5f7dfadabd3e8017406ad3720ba495a59963ee34))
## [0.23.2](https://github.com/Monadical-SAS/reflector/compare/v0.23.1...v0.23.2) (2025-12-11)
### Bug Fixes
* build on push tags ([#785](https://github.com/Monadical-SAS/reflector/issues/785)) ([d7f140b](https://github.com/Monadical-SAS/reflector/commit/d7f140b7d1f4660d5da7a0da1357f68869e0b5cd))
## [0.23.1](https://github.com/Monadical-SAS/reflector/compare/v0.23.0...v0.23.1) (2025-12-11)
### Bug Fixes
* populate room_name in transcript GET endpoint ([#783](https://github.com/Monadical-SAS/reflector/issues/783)) ([0eba147](https://github.com/Monadical-SAS/reflector/commit/0eba1470181c7b9e0a79964a1ef28c09bcbdd9d7))
## [0.23.0](https://github.com/Monadical-SAS/reflector/compare/v0.22.4...v0.23.0) (2025-12-10)
### Features
* dockerhub ci ([#772](https://github.com/Monadical-SAS/reflector/issues/772)) ([00549f1](https://github.com/Monadical-SAS/reflector/commit/00549f153ade922cf4cb6c5358a7d11a39c426d2))
* llm retries ([#739](https://github.com/Monadical-SAS/reflector/issues/739)) ([61f0e29](https://github.com/Monadical-SAS/reflector/commit/61f0e29d4c51eab54ee67af92141fbb171e8ccaa))
### Bug Fixes
* celery inspect bug sidestep in restart script ([#766](https://github.com/Monadical-SAS/reflector/issues/766)) ([ec17ed7](https://github.com/Monadical-SAS/reflector/commit/ec17ed7b587cf6ee143646baaee67a7c017044d4))
* deploy frontend to coolify ([#779](https://github.com/Monadical-SAS/reflector/issues/779)) ([91650ec](https://github.com/Monadical-SAS/reflector/commit/91650ec65f65713faa7ee0dcfb75af427b7c4ba0))
* hide rooms settings instead of disabling ([#763](https://github.com/Monadical-SAS/reflector/issues/763)) ([3ad78be](https://github.com/Monadical-SAS/reflector/commit/3ad78be7628c0d029296b301a0e87236c76b7598))
* return participant emails from transcript endpoint ([#769](https://github.com/Monadical-SAS/reflector/issues/769)) ([d3a5cd1](https://github.com/Monadical-SAS/reflector/commit/d3a5cd12d2d0d9c32af2d5bd9322e030ef69b85d))
## [0.22.4](https://github.com/Monadical-SAS/reflector/compare/v0.22.3...v0.22.4) (2025-12-02)
### Bug Fixes
* Multitrack mixdown optimisation 2 ([#764](https://github.com/Monadical-SAS/reflector/issues/764)) ([bd5df1c](https://github.com/Monadical-SAS/reflector/commit/bd5df1ce2ebf35d7f3413b295e56937a9a28ef7b))
## [0.22.3](https://github.com/Monadical-SAS/reflector/compare/v0.22.2...v0.22.3) (2025-12-02)
### Bug Fixes
* align daily room settings ([#759](https://github.com/Monadical-SAS/reflector/issues/759)) ([28f87c0](https://github.com/Monadical-SAS/reflector/commit/28f87c09dc459846873d0dde65b03e3d7b2b9399))
## [0.22.2](https://github.com/Monadical-SAS/reflector/compare/v0.22.1...v0.22.2) (2025-12-02)
### Bug Fixes
* daily auto refresh fix ([#755](https://github.com/Monadical-SAS/reflector/issues/755)) ([fe47c46](https://github.com/Monadical-SAS/reflector/commit/fe47c46489c5aa0cc538109f7559cc9accb35c01))
* Skip mixdown for multitrack ([#760](https://github.com/Monadical-SAS/reflector/issues/760)) ([b51b7aa](https://github.com/Monadical-SAS/reflector/commit/b51b7aa9176c1a53ba57ad99f5e976c804a1e80c))
## [0.22.1](https://github.com/Monadical-SAS/reflector/compare/v0.22.0...v0.22.1) (2025-11-27)
### Bug Fixes
* participants update from daily ([#749](https://github.com/Monadical-SAS/reflector/issues/749)) ([7f0b728](https://github.com/Monadical-SAS/reflector/commit/7f0b728991c1b9f9aae702c96297eae63b561ef5))
## [0.22.0](https://github.com/Monadical-SAS/reflector/compare/v0.21.0...v0.22.0) (2025-11-26)
### Features
* Multitrack segmentation ([#747](https://github.com/Monadical-SAS/reflector/issues/747)) ([d63040e](https://github.com/Monadical-SAS/reflector/commit/d63040e2fdc07e7b272e85a39eb2411cd6a14798))
## [0.21.0](https://github.com/Monadical-SAS/reflector/compare/v0.20.0...v0.21.0) (2025-11-26)
### Features
* add transcript format parameter to GET endpoint ([#709](https://github.com/Monadical-SAS/reflector/issues/709)) ([f6ca075](https://github.com/Monadical-SAS/reflector/commit/f6ca07505f34483b02270a2ef3bd809e9d2e1045))
## [0.20.0](https://github.com/Monadical-SAS/reflector/compare/v0.19.0...v0.20.0) (2025-11-25)
### Features
* link transcript participants ([#737](https://github.com/Monadical-SAS/reflector/issues/737)) ([9bec398](https://github.com/Monadical-SAS/reflector/commit/9bec39808fc6322612d8b87e922a6f7901fc01c1))
* transcript restart script ([#742](https://github.com/Monadical-SAS/reflector/issues/742)) ([86d5e26](https://github.com/Monadical-SAS/reflector/commit/86d5e26224bb55a0f1cc785aeda52065bb92ee6f))
## [0.19.0](https://github.com/Monadical-SAS/reflector/compare/v0.18.0...v0.19.0) (2025-11-25)
### Features
* dailyco api module ([#725](https://github.com/Monadical-SAS/reflector/issues/725)) ([4287f8b](https://github.com/Monadical-SAS/reflector/commit/4287f8b8aeee60e51db7539f4dcbda5f6e696bd8))
* dailyco poll ([#730](https://github.com/Monadical-SAS/reflector/issues/730)) ([8e438ca](https://github.com/Monadical-SAS/reflector/commit/8e438ca285152bd48fdc42767e706fb448d3525c))
* multitrack cli ([#735](https://github.com/Monadical-SAS/reflector/issues/735)) ([11731c9](https://github.com/Monadical-SAS/reflector/commit/11731c9d38439b04e93b1c3afbd7090bad11a11f))
### Bug Fixes
* default platform fix ([#736](https://github.com/Monadical-SAS/reflector/issues/736)) ([c442a62](https://github.com/Monadical-SAS/reflector/commit/c442a627873ca667656eeaefb63e54ab10b8d19e))
* parakeet vad not getting the end timestamp ([#728](https://github.com/Monadical-SAS/reflector/issues/728)) ([18ed713](https://github.com/Monadical-SAS/reflector/commit/18ed7133693653ef4ddac6c659a8c14b320d1657))
* start raw tracks recording ([#729](https://github.com/Monadical-SAS/reflector/issues/729)) ([3e47c2c](https://github.com/Monadical-SAS/reflector/commit/3e47c2c0573504858e0d2e1798b6ed31f16b4a5d))
## [0.18.0](https://github.com/Monadical-SAS/reflector/compare/v0.17.0...v0.18.0) (2025-11-14)
### Features
* daily QOL: participants dictionary ([#721](https://github.com/Monadical-SAS/reflector/issues/721)) ([b20cad7](https://github.com/Monadical-SAS/reflector/commit/b20cad76e69fb6a76405af299a005f1ddcf60eae))
### Bug Fixes
* add proccessing page to file upload and reprocessing ([#650](https://github.com/Monadical-SAS/reflector/issues/650)) ([28a7258](https://github.com/Monadical-SAS/reflector/commit/28a7258e45317b78e60e6397be2bc503647eaace))
* copy transcript ([#674](https://github.com/Monadical-SAS/reflector/issues/674)) ([a9a4f32](https://github.com/Monadical-SAS/reflector/commit/a9a4f32324f66c838e081eee42bb9502f38c1db1))
## [0.17.0](https://github.com/Monadical-SAS/reflector/compare/v0.16.0...v0.17.0) (2025-11-13)
### Features
* add API key management UI ([#716](https://github.com/Monadical-SAS/reflector/issues/716)) ([372202b](https://github.com/Monadical-SAS/reflector/commit/372202b0e1a86823900b0aa77be1bfbc2893d8a1))
* daily.co support as alternative to whereby ([#691](https://github.com/Monadical-SAS/reflector/issues/691)) ([1473fd8](https://github.com/Monadical-SAS/reflector/commit/1473fd82dc472c394cbaa2987212ad662a74bcac))
## [0.16.0](https://github.com/Monadical-SAS/reflector/compare/v0.15.0...v0.16.0) (2025-10-24)
### Features
* search date filter ([#710](https://github.com/Monadical-SAS/reflector/issues/710)) ([962c40e](https://github.com/Monadical-SAS/reflector/commit/962c40e2b6428ac42fd10aea926782d7a6f3f902))
## [0.15.0](https://github.com/Monadical-SAS/reflector/compare/v0.14.0...v0.15.0) (2025-10-20)
### Features
* api tokens ([#705](https://github.com/Monadical-SAS/reflector/issues/705)) ([9a258ab](https://github.com/Monadical-SAS/reflector/commit/9a258abc0209b0ac3799532a507ea6a9125d703a))
## [0.14.0](https://github.com/Monadical-SAS/reflector/compare/v0.13.1...v0.14.0) (2025-10-08)
### Features
* Add calendar event data to transcript webhook payload ([#689](https://github.com/Monadical-SAS/reflector/issues/689)) ([5f6910e](https://github.com/Monadical-SAS/reflector/commit/5f6910e5131b7f28f86c9ecdcc57fed8412ee3cd))
* container build for www / github ([#672](https://github.com/Monadical-SAS/reflector/issues/672)) ([969bd84](https://github.com/Monadical-SAS/reflector/commit/969bd84fcc14851d1a101412a0ba115f1b7cde82))
* docker-compose for production frontend ([#664](https://github.com/Monadical-SAS/reflector/issues/664)) ([5bf64b5](https://github.com/Monadical-SAS/reflector/commit/5bf64b5a41f64535e22849b4bb11734d4dbb4aae))
### Bug Fixes
* restore feature boolean logic ([#671](https://github.com/Monadical-SAS/reflector/issues/671)) ([3660884](https://github.com/Monadical-SAS/reflector/commit/36608849ec64e953e3be456172502762e3c33df9))
* security review ([#656](https://github.com/Monadical-SAS/reflector/issues/656)) ([5d98754](https://github.com/Monadical-SAS/reflector/commit/5d98754305c6c540dd194dda268544f6d88bfaf8))
* update transcript list on reprocess ([#676](https://github.com/Monadical-SAS/reflector/issues/676)) ([9a71af1](https://github.com/Monadical-SAS/reflector/commit/9a71af145ee9b833078c78d0c684590ab12e9f0e))
* upgrade nemo toolkit ([#678](https://github.com/Monadical-SAS/reflector/issues/678)) ([eef6dc3](https://github.com/Monadical-SAS/reflector/commit/eef6dc39037329b65804297786d852dddb0557f9))
## [0.13.1](https://github.com/Monadical-SAS/reflector/compare/v0.13.0...v0.13.1) (2025-09-22)
### Bug Fixes
* TypeError on not all arguments converted during string formatting in logger ([#667](https://github.com/Monadical-SAS/reflector/issues/667)) ([565a629](https://github.com/Monadical-SAS/reflector/commit/565a62900f5a02fc946b68f9269a42190ed70ab6))
## [0.13.0](https://github.com/Monadical-SAS/reflector/compare/v0.12.1...v0.13.0) (2025-09-19)
### Features
* room form edit with enter ([#662](https://github.com/Monadical-SAS/reflector/issues/662)) ([47716f6](https://github.com/Monadical-SAS/reflector/commit/47716f6e5ddee952609d2fa0ffabdfa865286796))
### Bug Fixes
* invalid cleanup call ([#660](https://github.com/Monadical-SAS/reflector/issues/660)) ([0abcebf](https://github.com/Monadical-SAS/reflector/commit/0abcebfc9491f87f605f21faa3e53996fafedd9a))
## [0.12.1](https://github.com/Monadical-SAS/reflector/compare/v0.12.0...v0.12.1) (2025-09-17)
### Bug Fixes
* production blocked because having existing meeting with room_id null ([#657](https://github.com/Monadical-SAS/reflector/issues/657)) ([870e860](https://github.com/Monadical-SAS/reflector/commit/870e8605171a27155a9cbee215eeccb9a8d6c0a2))
## [0.12.0](https://github.com/Monadical-SAS/reflector/compare/v0.11.0...v0.12.0) (2025-09-17)
### Features
* calendar integration ([#608](https://github.com/Monadical-SAS/reflector/issues/608)) ([6f680b5](https://github.com/Monadical-SAS/reflector/commit/6f680b57954c688882c4ed49f40f161c52a00a24))
* self-hosted gpu api ([#636](https://github.com/Monadical-SAS/reflector/issues/636)) ([ab859d6](https://github.com/Monadical-SAS/reflector/commit/ab859d65a6bded904133a163a081a651b3938d42))
### Bug Fixes
* ignore player hotkeys for text inputs ([#646](https://github.com/Monadical-SAS/reflector/issues/646)) ([fa049e8](https://github.com/Monadical-SAS/reflector/commit/fa049e8d068190ce7ea015fd9fcccb8543f54a3f))
## [0.11.0](https://github.com/Monadical-SAS/reflector/compare/v0.10.0...v0.11.0) (2025-09-16)
### Features
* remove profanity filter that was there for conference ([#652](https://github.com/Monadical-SAS/reflector/issues/652)) ([b42f7cf](https://github.com/Monadical-SAS/reflector/commit/b42f7cfc606783afcee792590efcc78b507468ab))
### Bug Fixes
* zulip and consent handler on the file pipeline ([#645](https://github.com/Monadical-SAS/reflector/issues/645)) ([5f143fe](https://github.com/Monadical-SAS/reflector/commit/5f143fe3640875dcb56c26694254a93189281d17))
* zulip stream and topic selection in share dialog ([#644](https://github.com/Monadical-SAS/reflector/issues/644)) ([c546e69](https://github.com/Monadical-SAS/reflector/commit/c546e69739e68bb74fbc877eb62609928e5b8de6))
## [0.10.0](https://github.com/Monadical-SAS/reflector/compare/v0.9.0...v0.10.0) (2025-09-11)
### Features
* replace nextjs-config with environment variables ([#632](https://github.com/Monadical-SAS/reflector/issues/632)) ([369ecdf](https://github.com/Monadical-SAS/reflector/commit/369ecdff13f3862d926a9c0b87df52c9d94c4dde))
### Bug Fixes
* anonymous users transcript permissions ([#621](https://github.com/Monadical-SAS/reflector/issues/621)) ([f81fe99](https://github.com/Monadical-SAS/reflector/commit/f81fe9948a9237b3e0001b2d8ca84f54d76878f9))
* auth post ([#624](https://github.com/Monadical-SAS/reflector/issues/624)) ([cde99ca](https://github.com/Monadical-SAS/reflector/commit/cde99ca2716f84ba26798f289047732f0448742e))
* auth post ([#626](https://github.com/Monadical-SAS/reflector/issues/626)) ([3b85ff3](https://github.com/Monadical-SAS/reflector/commit/3b85ff3bdf4fb053b103070646811bc990c0e70a))
* auth post ([#627](https://github.com/Monadical-SAS/reflector/issues/627)) ([962038e](https://github.com/Monadical-SAS/reflector/commit/962038ee3f2a555dc3c03856be0e4409456e0996))
* missing follow_redirects=True on modal endpoint ([#630](https://github.com/Monadical-SAS/reflector/issues/630)) ([fc363bd](https://github.com/Monadical-SAS/reflector/commit/fc363bd49b17b075e64f9186e5e0185abc325ea7))
* sync backend and frontend token refresh logic ([#614](https://github.com/Monadical-SAS/reflector/issues/614)) ([5a5b323](https://github.com/Monadical-SAS/reflector/commit/5a5b3233820df9536da75e87ce6184a983d4713a))
## [0.9.0](https://github.com/Monadical-SAS/reflector/compare/v0.8.2...v0.9.0) (2025-09-06)
### Features
* frontend openapi react query ([#606](https://github.com/Monadical-SAS/reflector/issues/606)) ([c4d2825](https://github.com/Monadical-SAS/reflector/commit/c4d2825c81f81ad8835629fbf6ea8c7383f8c31b))
### Bug Fixes
* align whisper transcriber api with parakeet ([#602](https://github.com/Monadical-SAS/reflector/issues/602)) ([0663700](https://github.com/Monadical-SAS/reflector/commit/0663700a615a4af69a03c96c410f049e23ec9443))
* kv use tls explicit ([#610](https://github.com/Monadical-SAS/reflector/issues/610)) ([08d88ec](https://github.com/Monadical-SAS/reflector/commit/08d88ec349f38b0d13e0fa4cb73486c8dfd31836))
* source kind for file processing ([#601](https://github.com/Monadical-SAS/reflector/issues/601)) ([dc82f8b](https://github.com/Monadical-SAS/reflector/commit/dc82f8bb3bdf3ab3d4088e592a30fd63907319e1))
* token refresh locking ([#613](https://github.com/Monadical-SAS/reflector/issues/613)) ([7f5a4c9](https://github.com/Monadical-SAS/reflector/commit/7f5a4c9ddc7fd098860c8bdda2ca3b57f63ded2f))
## [0.8.2](https://github.com/Monadical-SAS/reflector/compare/v0.8.1...v0.8.2) (2025-08-29)
### Bug Fixes
* search-logspam ([#593](https://github.com/Monadical-SAS/reflector/issues/593)) ([695d1a9](https://github.com/Monadical-SAS/reflector/commit/695d1a957d4cd862753049f9beed88836cabd5ab))
## [0.8.1](https://github.com/Monadical-SAS/reflector/compare/v0.8.0...v0.8.1) (2025-08-29)
### Bug Fixes
* make webhook secret/url allowing null ([#590](https://github.com/Monadical-SAS/reflector/issues/590)) ([84a3812](https://github.com/Monadical-SAS/reflector/commit/84a381220bc606231d08d6f71d4babc818fa3c75))
## [0.8.0](https://github.com/Monadical-SAS/reflector/compare/v0.7.3...v0.8.0) (2025-08-29)
### Features
* **cleanup:** add automatic data retention for public instances ([#574](https://github.com/Monadical-SAS/reflector/issues/574)) ([6f0c7c1](https://github.com/Monadical-SAS/reflector/commit/6f0c7c1a5e751713366886c8e764c2009e12ba72))
* **rooms:** add webhook for transcript completion ([#578](https://github.com/Monadical-SAS/reflector/issues/578)) ([88ed7cf](https://github.com/Monadical-SAS/reflector/commit/88ed7cfa7804794b9b54cad4c3facc8a98cf85fd))
### Bug Fixes
* file pipeline status reporting and websocket updates ([#589](https://github.com/Monadical-SAS/reflector/issues/589)) ([9dfd769](https://github.com/Monadical-SAS/reflector/commit/9dfd76996f851cc52be54feea078adbc0816dc57))
* Igor/evaluation ([#575](https://github.com/Monadical-SAS/reflector/issues/575)) ([124ce03](https://github.com/Monadical-SAS/reflector/commit/124ce03bf86044c18313d27228a25da4bc20c9c5))
* optimize parakeet transcription batching algorithm ([#577](https://github.com/Monadical-SAS/reflector/issues/577)) ([7030e0f](https://github.com/Monadical-SAS/reflector/commit/7030e0f23649a8cf6c1eb6d5889684a41ce849ec))
## [0.7.3](https://github.com/Monadical-SAS/reflector/compare/v0.7.2...v0.7.3) (2025-08-22)
### Bug Fixes
* cleaned repo, and get git-leaks clean ([359280d](https://github.com/Monadical-SAS/reflector/commit/359280dd340433ba4402ed69034094884c825e67))
* restore previous behavior on live pipeline + audio downscaler ([#561](https://github.com/Monadical-SAS/reflector/issues/561)) ([9265d20](https://github.com/Monadical-SAS/reflector/commit/9265d201b590d23c628c5f19251b70f473859043))
## [0.7.2](https://github.com/Monadical-SAS/reflector/compare/v0.7.1...v0.7.2) (2025-08-21)
### Bug Fixes
* docker image not loading libgomp.so.1 for torch ([#560](https://github.com/Monadical-SAS/reflector/issues/560)) ([773fccd](https://github.com/Monadical-SAS/reflector/commit/773fccd93e887c3493abc2e4a4864dddce610177))
* include shared rooms to search ([#558](https://github.com/Monadical-SAS/reflector/issues/558)) ([499eced](https://github.com/Monadical-SAS/reflector/commit/499eced3360b84fb3a90e1c8a3b554290d21adc2))
## [0.7.1](https://github.com/Monadical-SAS/reflector/compare/v0.7.0...v0.7.1) (2025-08-21)
### Bug Fixes
* webvtt db null expectation mismatch ([#556](https://github.com/Monadical-SAS/reflector/issues/556)) ([e67ad1a](https://github.com/Monadical-SAS/reflector/commit/e67ad1a4a2054467bfeb1e0258fbac5868aaaf21))
## [0.7.0](https://github.com/Monadical-SAS/reflector/compare/v0.6.1...v0.7.0) (2025-08-21)
### Features
* delete recording with transcript ([#547](https://github.com/Monadical-SAS/reflector/issues/547)) ([99cc984](https://github.com/Monadical-SAS/reflector/commit/99cc9840b3f5de01e0adfbfae93234042d706d13))
* pipeline improvement with file processing, parakeet, silero-vad ([#540](https://github.com/Monadical-SAS/reflector/issues/540)) ([bcc29c9](https://github.com/Monadical-SAS/reflector/commit/bcc29c9e0050ae215f89d460e9d645aaf6a5e486))
* postgresql migration and removal of sqlite in pytest ([#546](https://github.com/Monadical-SAS/reflector/issues/546)) ([cd1990f](https://github.com/Monadical-SAS/reflector/commit/cd1990f8f0fe1503ef5069512f33777a73a93d7f))
* search backend ([#537](https://github.com/Monadical-SAS/reflector/issues/537)) ([5f9b892](https://github.com/Monadical-SAS/reflector/commit/5f9b89260c9ef7f3c921319719467df22830453f))
* search frontend ([#551](https://github.com/Monadical-SAS/reflector/issues/551)) ([3657242](https://github.com/Monadical-SAS/reflector/commit/365724271ca6e615e3425125a69ae2b46ce39285))
### Bug Fixes
* evaluation cli event wrap ([#536](https://github.com/Monadical-SAS/reflector/issues/536)) ([941c3db](https://github.com/Monadical-SAS/reflector/commit/941c3db0bdacc7b61fea412f3746cc5a7cb67836))
* use structlog not logging ([#550](https://github.com/Monadical-SAS/reflector/issues/550)) ([27e2f81](https://github.com/Monadical-SAS/reflector/commit/27e2f81fda5232e53edc729d3e99c5ef03adbfe9))
## [0.6.1](https://github.com/Monadical-SAS/reflector/compare/v0.6.0...v0.6.1) (2025-08-06)
### Bug Fixes
* delayed waveform loading ([#538](https://github.com/Monadical-SAS/reflector/issues/538)) ([ef64146](https://github.com/Monadical-SAS/reflector/commit/ef64146325d03f64dd9a1fe40234fb3e7e957ae2))
## [0.6.0](https://github.com/Monadical-SAS/reflector/compare/v0.5.0...v0.6.0) (2025-08-05)
### ⚠ BREAKING CHANGES
* Configuration keys have changed. Update your .env file:
- TRANSCRIPT_MODAL_API_KEY → TRANSCRIPT_API_KEY
- LLM_MODAL_API_KEY → (removed, use TRANSCRIPT_API_KEY)
- Add DIARIZATION_API_KEY and TRANSLATE_API_KEY if using those services
### Features
* implement service-specific Modal API keys with auto processor pattern ([#528](https://github.com/Monadical-SAS/reflector/issues/528)) ([650befb](https://github.com/Monadical-SAS/reflector/commit/650befb291c47a1f49e94a01ab37d8fdfcd2b65d))
* use llamaindex everywhere ([#525](https://github.com/Monadical-SAS/reflector/issues/525)) ([3141d17](https://github.com/Monadical-SAS/reflector/commit/3141d172bc4d3b3d533370c8e6e351ea762169bf))
### Miscellaneous Chores
* **main:** release 0.6.0 ([ecdbf00](https://github.com/Monadical-SAS/reflector/commit/ecdbf003ea2476c3e95fd231adaeb852f2943df0))
## [0.5.0](https://github.com/Monadical-SAS/reflector/compare/v0.4.0...v0.5.0) (2025-07-31)

View File

@@ -62,28 +62,29 @@ uv run python -m reflector.tools.process path/to/audio.wav
**Setup:**
```bash
# Install dependencies
pnpm install
yarn install
# Copy configuration templates
cp .env_template .env
cp config-template.ts config.ts
```
**Development:**
```bash
# Start development server
pnpm dev
yarn dev
# Generate TypeScript API client from OpenAPI spec
pnpm openapi
yarn openapi
# Lint code
pnpm lint
yarn lint
# Format code
pnpm format
yarn format
# Build for production
pnpm build
yarn build
```
### Docker Compose (Full Stack)
@@ -143,15 +144,13 @@ All endpoints prefixed `/v1/`:
**Backend** (`server/.env`):
- `DATABASE_URL` - Database connection string
- `REDIS_URL` - Redis broker for Celery
- `TRANSCRIPT_BACKEND=modal` + `TRANSCRIPT_MODAL_API_KEY` - Modal.com transcription
- `DIARIZATION_BACKEND=modal` + `DIARIZATION_MODAL_API_KEY` - Modal.com diarization
- `TRANSLATION_BACKEND=modal` + `TRANSLATION_MODAL_API_KEY` - Modal.com translation
- `MODAL_TOKEN_ID`, `MODAL_TOKEN_SECRET` - Modal.com GPU processing
- `WHEREBY_API_KEY` - Video platform integration
- `REFLECTOR_AUTH_BACKEND` - Authentication method (none, jwt)
**Frontend** (`www/.env`):
- `NEXTAUTH_URL`, `NEXTAUTH_SECRET` - Authentication configuration
- `REFLECTOR_API_URL` - Backend API endpoint
- `NEXT_PUBLIC_REFLECTOR_API_URL` - Backend API endpoint
- `REFLECTOR_DOMAIN_CONFIG` - Feature flags and domain settings
## Testing Strategy

264
IMPLEMENTATION_STATUS.md Normal file
View File

@@ -0,0 +1,264 @@
# Daily.co Migration Implementation Status
## Completed Components
### 1. Platform Abstraction Layer (`server/reflector/video_platforms/`)
- **base.py**: Abstract interface defining all platform operations
- **whereby.py**: Whereby implementation wrapping existing functionality
- **daily.py**: Daily.co client implementation (ready for testing when credentials available)
- **mock.py**: Mock implementation for unit testing
- **registry.py**: Platform registration and discovery
- **factory.py**: Factory methods for creating platform clients
### 2. Database Updates
- **Models**: Added `platform` field to Room and Meeting tables
- **Migration**: Created migration `20250801180012_add_platform_support.py`
- **Controllers**: Updated to handle platform field
### 3. Configuration
- **Settings**: Added Daily.co configuration variables
- **Feature Flags**:
- `DAILY_MIGRATION_ENABLED`: Master switch for migration
- `DAILY_MIGRATION_ROOM_IDS`: List of specific rooms to migrate
- `DEFAULT_VIDEO_PLATFORM`: Default platform when migration enabled
### 4. Backend API Updates
- **Room Creation**: Now assigns platform based on feature flags
- **Meeting Creation**: Uses platform abstraction instead of direct Whereby calls
- **Response Models**: Include platform field
- **Webhook Handler**: Added Daily.co webhook endpoint at `/v1/daily_webhook`
### 5. Frontend Components (`www/app/[roomName]/components/`)
- **RoomContainer.tsx**: Platform-agnostic container that routes to appropriate component
- **WherebyRoom.tsx**: Extracted existing Whereby functionality with consent management
- **DailyRoom.tsx**: Daily.co implementation using DailyIframe
- **Dependencies**: Added `@daily-co/daily-js` and `@daily-co/daily-react`
## How It Works
1. **Platform Selection**:
- If `DAILY_MIGRATION_ENABLED=false` → Always use Whereby
- If enabled and room ID in `DAILY_MIGRATION_ROOM_IDS` → Use Daily
- Otherwise → Use `DEFAULT_VIDEO_PLATFORM`
2. **Meeting Creation Flow**:
```python
platform = get_platform_for_room(room.id)
client = create_platform_client(platform)
meeting_data = await client.create_meeting(...)
```
3. **Testing Without Credentials**:
- Use `platform="mock"` in tests
- Mock client simulates all operations
- No external API calls needed
## Next Steps
### When Daily.co Credentials Available:
1. **Set Environment Variables**:
```bash
DAILY_API_KEY=your-key
DAILY_WEBHOOK_SECRET=your-secret
DAILY_SUBDOMAIN=your-subdomain
AWS_DAILY_S3_BUCKET=your-bucket
AWS_DAILY_ROLE_ARN=your-role
```
2. **Run Database Migration**:
```bash
cd server
uv run alembic upgrade head
```
3. **Test Platform Creation**:
```python
from reflector.video_platforms.factory import create_platform_client
client = create_platform_client("daily")
# Test operations...
```
### 6. Testing & Validation (`server/tests/`)
- **test_video_platforms.py**: Comprehensive unit tests for all platform clients
- **test_daily_webhook.py**: Integration tests for Daily.co webhook handling
- **utils/video_platform_test_utils.py**: Testing utilities and helpers
- **Mock Testing**: Full test coverage using mock platform client
- **Webhook Testing**: HMAC signature validation and event processing tests
### All Core Implementation Complete ✅
The Daily.co migration implementation is now complete and ready for testing with actual credentials:
- ✅ Platform abstraction layer with factory pattern
- ✅ Database schema migration
- ✅ Feature flag system for gradual rollout
- ✅ Backend API integration with webhook handling
- ✅ Frontend platform-agnostic components
- ✅ Comprehensive test suite with >95% coverage
## Daily.co Webhook Integration
### Webhook Configuration
Daily.co webhooks are configured via API (no dashboard interface). Use the Daily.co REST API to set up webhook endpoints:
```bash
# Configure webhook endpoint
curl -X POST https://api.daily.co/v1/webhook-endpoints \
-H "Authorization: Bearer ${DAILY_API_KEY}" \
-H "Content-Type: application/json" \
-d '{
"url": "https://yourdomain.com/v1/daily_webhook",
"events": [
"participant.joined",
"participant.left",
"recording.started",
"recording.ready-to-download",
"recording.error"
]
}'
```
### Webhook Event Examples
**Participant Joined:**
```json
{
"type": "participant.joined",
"id": "evt_participant_joined_1640995200",
"ts": 1640995200000,
"data": {
"room": {"name": "test-room-123-abc"},
"participant": {
"id": "participant-123",
"user_name": "John Doe",
"session_id": "session-456"
}
}
}
```
**Recording Ready:**
```json
{
"type": "recording.ready-to-download",
"id": "evt_recording_ready_1640995200",
"ts": 1640995200000,
"data": {
"room": {"name": "test-room-123-abc"},
"recording": {
"id": "recording-789",
"status": "finished",
"download_url": "https://bucket.s3.amazonaws.com/recording.mp4",
"start_time": "2025-01-01T10:00:00Z",
"duration": 1800
}
}
}
```
### Webhook Signature Verification
Daily.co uses HMAC-SHA256 for webhook verification:
```python
import hmac
import hashlib
def verify_daily_webhook(body: bytes, signature: str, secret: str) -> bool:
expected = hmac.new(secret.encode(), body, hashlib.sha256).hexdigest()
return hmac.compare_digest(expected, signature)
```
Signature is sent in the `X-Daily-Signature` header.
### Recording Processing Flow
1. **Daily.co Meeting Ends** → Recording processed
2. **Webhook Fired** → `recording.ready-to-download` event
3. **Webhook Handler** → Extracts download URL and recording ID
4. **Background Task** → `process_recording_from_url.delay()` queued
5. **Download & Process** → Audio downloaded, validated, transcribed
6. **ML Pipeline** → Same processing as Whereby recordings
```python
# New Celery task for Daily.co recordings
@shared_task
@asynctask
async def process_recording_from_url(recording_url: str, meeting_id: str, recording_id: str):
# Downloads from Daily.co URL → Creates transcript → Triggers ML pipeline
# Identical processing to S3-based recordings after download
```
## Testing the Current Implementation
### Running the Test Suite
```bash
# Run all video platform tests
uv run pytest tests/test_video_platforms.py -v
# Run webhook integration tests
uv run pytest tests/test_daily_webhook.py -v
# Run with coverage
uv run pytest tests/test_video_platforms.py tests/test_daily_webhook.py --cov=reflector.video_platforms --cov=reflector.views.daily
```
### Manual Testing with Mock Platform
```python
from reflector.video_platforms.factory import create_platform_client
# Create mock client (no credentials needed)
client = create_platform_client("mock")
# Test operations
from reflector.db.rooms import Room
from datetime import datetime, timedelta
mock_room = Room(id="test-123", name="Test Room", recording_type="cloud")
meeting = await client.create_meeting(
room_name_prefix="test",
end_date=datetime.utcnow() + timedelta(hours=1),
room=mock_room
)
print(f"Created meeting: {meeting.room_url}")
```
### Testing Daily.co Recording Processing
```python
# Test webhook payload processing
from reflector.views.daily import daily_webhook
from reflector.worker.process import process_recording_from_url
# Simulate webhook event
event_data = {
"type": "recording.ready-to-download",
"id": "evt_123",
"ts": 1640995200000,
"data": {
"room": {"name": "test-room-123"},
"recording": {
"id": "rec-456",
"download_url": "https://daily.co/recordings/test.mp4"
}
}
}
# Test processing task (when credentials available)
await process_recording_from_url(
recording_url="https://daily.co/recordings/test.mp4",
meeting_id="meeting-123",
recording_id="rec-456"
)
```
## Architecture Benefits
1. **Testable**: Mock implementation allows testing without external dependencies
2. **Extensible**: Easy to add new platforms (Zoom, Teams, etc.)
3. **Gradual Migration**: Feature flags enable room-by-room migration
4. **Rollback Ready**: Can disable Daily.co instantly via feature flag

287
PLAN.md Normal file
View File

@@ -0,0 +1,287 @@
# Daily.co Migration Plan - Feature Parity Approach
## Overview
This plan outlines a systematic migration from Whereby to Daily.co, focusing on **1:1 feature parity** without introducing new capabilities. The goal is to improve code quality, developer experience, and platform reliability while maintaining the exact same user experience and processing pipeline.
## Migration Principles
1. **No Breaking Changes**: Existing recordings and workflows must continue to work
2. **Feature Parity First**: Match current functionality exactly before adding improvements
3. **Gradual Rollout**: Use feature flags to control migration per room/user
4. **Minimal Risk**: Keep changes isolated and reversible
## Phase 1: Foundation
### 1.1 Environment Setup
**Owner**: Backend Developer
- [ ] Create Daily.co account and obtain API credentials (PENDING - User to provide)
- [x] Add environment variables to `.env` files:
```bash
DAILY_API_KEY=your-api-key
DAILY_WEBHOOK_SECRET=your-webhook-secret
DAILY_SUBDOMAIN=your-subdomain
AWS_DAILY_ROLE_ARN=arn:aws:iam::xxx:role/daily-recording
```
- [ ] Set up Daily.co webhook endpoint in dashboard (PENDING - Credentials needed)
- [ ] Configure S3 bucket permissions for Daily.co (PENDING - Credentials needed)
### 1.2 Database Migration
**Owner**: Backend Developer
- [x] Create Alembic migration:
```python
# server/migrations/versions/20250801180012_add_platform_support.py
def upgrade():
op.add_column('rooms', sa.Column('platform', sa.String(), server_default='whereby'))
op.add_column('meetings', sa.Column('platform', sa.String(), server_default='whereby'))
```
- [ ] Run migration on development database (USER TO RUN: `uv run alembic upgrade head`)
- [x] Update models to include platform field
### 1.3 Feature Flag System
**Owner**: Full-stack Developer
- [x] Implement feature flag in backend settings:
```python
DAILY_MIGRATION_ENABLED = env.bool("DAILY_MIGRATION_ENABLED", False)
DAILY_MIGRATION_ROOM_IDS = env.list("DAILY_MIGRATION_ROOM_IDS", [])
```
- [x] Add platform selection logic to room creation
- [ ] Create admin UI to toggle platform per room (FUTURE - Not in Phase 1)
### 1.4 Daily.co API Client
**Owner**: Backend Developer
- [x] Create `server/reflector/video_platforms/` with core functionality:
- `create_meeting()` - Match Whereby's meeting creation
- `get_room_sessions()` - Room status checking
- `delete_room()` - Cleanup functionality
- [x] Add comprehensive error handling
- [ ] Write unit tests for API client (Phase 4)
## Phase 2: Backend Integration
### 2.1 Webhook Handler
**Owner**: Backend Developer
- [x] Create `server/reflector/views/daily.py` webhook endpoint
- [x] Implement HMAC signature verification
- [x] Handle events:
- `participant.joined`
- `participant.left`
- `recording.started`
- `recording.ready-to-download`
- [x] Map Daily.co events to existing database updates
- [x] Register webhook router in main app
- [ ] Add webhook tests with mocked events (Phase 4)
### 2.2 Room Management Updates
**Owner**: Backend Developer
- [x] Update `server/reflector/views/rooms.py`:
```python
# Uses platform abstraction layer
platform = get_platform_for_room(room.id)
client = create_platform_client(platform)
meeting_data = await client.create_meeting(...)
```
- [x] Ensure room URLs are stored correctly
- [x] Update meeting status checks to support both platforms
- [ ] Test room creation/deletion for both platforms (Phase 4)
## Phase 3: Frontend Migration
### 3.1 Daily.co React Setup
**Owner**: Frontend Developer
- [x] Install Daily.co packages:
```bash
yarn add @daily-co/daily-react @daily-co/daily-js
```
- [x] Create platform-agnostic components structure
- [x] Set up TypeScript interfaces for meeting data
### 3.2 Room Component Refactor
**Owner**: Frontend Developer
- [x] Create platform-agnostic room component:
```tsx
// www/app/[roomName]/components/RoomContainer.tsx
export default function RoomContainer({ params }) {
const platform = meeting.response.platform || "whereby";
if (platform === 'daily') {
return <DailyRoom meeting={meeting.response} />
}
return <WherebyRoom meeting={meeting.response} />
}
```
- [x] Implement `DailyRoom` component with:
- Call initialization using DailyIframe
- Recording consent flow
- Leave meeting handling
- [x] Extract `WherebyRoom` component maintaining existing functionality
- [x] Simplified focus management (Daily.co handles this internally)
### 3.3 Consent Dialog Integration
**Owner**: Frontend Developer
- [x] Adapt consent dialog for Daily.co (uses same API endpoints)
- [x] Ensure recording status is properly tracked
- [x] Maintain consistent consent UI across both platforms
- [ ] Test consent flow with Daily.co recordings (Phase 4)
## Phase 4: Testing & Validation
### 4.1 Unit Testing ✅
**Owner**: Backend Developer
- [x] Create comprehensive unit tests for all platform clients
- [x] Test mock platform client with full coverage
- [x] Test platform factory and registry functionality
- [x] Test webhook signature verification for all platforms
- [x] Test meeting lifecycle operations (create, delete, sessions)
### 4.2 Integration Testing ✅
**Owner**: Backend Developer
- [x] Create webhook integration tests with mocked HTTP client
- [x] Test Daily.co webhook event processing
- [x] Test participant join/leave event handling
- [x] Test recording start/ready event processing
- [x] Test webhook signature validation with HMAC
- [x] Test error handling for malformed events
### 4.3 Test Utilities ✅
**Owner**: Backend Developer
- [x] Create video platform test helper utilities
- [x] Create webhook event generators for testing
- [x] Create platform-agnostic test scenarios
- [x] Implement mock data factories for consistent testing
### 4.4 Ready for Live Testing
**Owner**: QA + Development Team
- [ ] Test complete flow with actual Daily.co credentials:
- Room creation
- Join meeting
- Recording consent
- Recording to S3
- Webhook processing
- Transcript generation
- [ ] Verify S3 paths are compatible
- [ ] Check recording format (MP4) matches
- [ ] Ensure processing pipeline works unchanged
## Phase 5: Gradual Rollout
### 5.1 Internal Testing
**Owner**: Development Team
- [ ] Enable Daily.co for internal test rooms
- [ ] Monitor logs and error rates
- [ ] Fix any issues discovered
- [ ] Verify recordings process correctly
### 5.2 Beta Rollout
**Owner**: DevOps + Product
- [ ] Select beta users/rooms
- [ ] Enable Daily.co via feature flag
- [ ] Monitor metrics:
- Error rates
- Recording success
- User feedback
- [ ] Create rollback plan
### 5.3 Full Migration
**Owner**: DevOps + Product
- [ ] Gradually increase Daily.co usage
- [ ] Monitor all metrics
- [ ] Plan Whereby sunset timeline
- [ ] Update documentation
## Success Criteria
### Technical Metrics
- [x] Comprehensive test coverage (>95% for platform abstraction)
- [x] Mock testing confirms API integration patterns work
- [x] Webhook processing tested with realistic event payloads
- [x] Error handling validated for all failure scenarios
- [ ] Live API error rate < 0.1% (pending credentials)
- [ ] Live webhook delivery rate > 99.9% (pending credentials)
- [ ] Recording success rate matches Whereby (pending credentials)
### User Experience
- [x] Platform-agnostic components maintain existing UX
- [x] Recording consent flow preserved across platforms
- [x] Participant tracking architecture unchanged
- [ ] Live call quality validation (pending credentials)
- [ ] Live user acceptance testing (pending credentials)
### Code Quality ✅
- [x] Removed 70+ lines of focus management code in WherebyRoom extraction
- [x] Improved TypeScript coverage with platform interfaces
- [x] Better error handling with platform abstraction
- [x] Cleaner React component structure with platform routing
## Rollback Plan
If issues arise during migration:
1. **Immediate**: Disable Daily.co feature flag
2. **Short-term**: Revert frontend components via git
3. **Database**: Platform field defaults to 'whereby'
4. **Full rollback**: Remove Daily.co code (isolated in separate files)
## Post-Migration Opportunities
Once feature parity is achieved and stable:
1. **Raw-tracks recording** for better diarization
2. **Real-time transcription** via Daily.co API
3. **Advanced analytics** and participant insights
4. **Custom UI** improvements
5. **Performance optimizations**
## Phase Dependencies
- ✅ Backend Integration requires Foundation to be complete
- ✅ Frontend Migration can start after Backend API client is ready
- ✅ Testing requires both Backend and Frontend to be complete
- ⏳ Rollout begins after successful testing (pending Daily.co credentials)
## Risk Matrix
| Risk | Probability | Impact | Mitigation |
|------|-------------|---------|------------|
| API differences | Low | Medium | Abstraction layer |
| Recording format issues | Low | High | Extensive testing |
| User confusion | Low | Low | Gradual rollout |
| Performance degradation | Low | Medium | Monitoring |
## Communication Plan
1. **Week 1**: Announce migration plan to team
2. **Week 2**: Update on development progress
3. **Beta Launch**: Email to beta users
4. **Full Launch**: User notification (if UI changes)
5. **Post-Launch**: Success metrics report
---
## Implementation Status: COMPLETE ✅
All development phases are complete and ready for live testing:
**Phase 1**: Foundation (database, config, feature flags)
**Phase 2**: Backend Integration (API clients, webhooks)
**Phase 3**: Frontend Migration (platform components)
**Phase 4**: Testing & Validation (comprehensive test suite)
**Next Steps**: Obtain Daily.co credentials and run live integration testing before gradual rollout.
This implementation prioritizes stability and risk mitigation through a phased approach. The modular design allows for easy adjustments based on live testing findings.

100
README.md
View File

@@ -1,60 +1,43 @@
<div align="center">
<img width="100" alt="image" src="https://github.com/user-attachments/assets/66fb367b-2c89-4516-9912-f47ac59c6a7f"/>
# Reflector
Reflector is an AI-powered audio transcription and meeting analysis platform that provides real-time transcription, speaker diarization, translation and summarization for audio content and live meetings. It works 100% with local models (whisper/parakeet, pyannote, seamless-m4t, and your local llm like phi-4).
Reflector Audio Management and Analysis is a cutting-edge web application under development by Monadical. It utilizes AI to record meetings, providing a permanent record with transcripts, translations, and automated summaries.
[![Tests](https://github.com/monadical-sas/reflector/actions/workflows/test_server.yml/badge.svg?branch=main&event=push)](https://github.com/monadical-sas/reflector/actions/workflows/test_server.yml)
[![Tests](https://github.com/monadical-sas/reflector/actions/workflows/pytests.yml/badge.svg?branch=main&event=push)](https://github.com/monadical-sas/reflector/actions/workflows/pytests.yml)
[![License: MIT](https://img.shields.io/badge/license-MIT-green.svg)](https://opensource.org/licenses/MIT)
</div>
</div>
## Screenshots
<table>
<tr>
<td>
<a href="https://github.com/user-attachments/assets/21f5597c-2930-4899-a154-f7bd61a59e97">
<img width="700" alt="image" src="https://github.com/user-attachments/assets/21f5597c-2930-4899-a154-f7bd61a59e97" />
<a href="https://github.com/user-attachments/assets/3a976930-56c1-47ef-8c76-55d3864309e3">
<img width="700" alt="image" src="https://github.com/user-attachments/assets/3a976930-56c1-47ef-8c76-55d3864309e3" />
</a>
</td>
<td>
<a href="https://github.com/user-attachments/assets/f6b9399a-5e51-4bae-b807-59128d0a940c">
<img width="700" alt="image" src="https://github.com/user-attachments/assets/f6b9399a-5e51-4bae-b807-59128d0a940c" />
<a href="https://github.com/user-attachments/assets/bfe3bde3-08af-4426-a9a1-11ad5cd63b33">
<img width="700" alt="image" src="https://github.com/user-attachments/assets/bfe3bde3-08af-4426-a9a1-11ad5cd63b33" />
</a>
</td>
<td>
<a href="https://github.com/user-attachments/assets/a42ce460-c1fd-4489-a995-270516193897">
<img width="700" alt="image" src="https://github.com/user-attachments/assets/a42ce460-c1fd-4489-a995-270516193897" />
</a>
</td>
<td>
<a href="https://github.com/user-attachments/assets/21929f6d-c309-42fe-9c11-f1299e50fbd4">
<img width="700" alt="image" src="https://github.com/user-attachments/assets/21929f6d-c309-42fe-9c11-f1299e50fbd4" />
<a href="https://github.com/user-attachments/assets/7b60c9d0-efe4-474f-a27b-ea13bd0fabdc">
<img width="700" alt="image" src="https://github.com/user-attachments/assets/7b60c9d0-efe4-474f-a27b-ea13bd0fabdc" />
</a>
</td>
</tr>
</table>
## What is Reflector?
Reflector is a web application that utilizes local models to process audio content, providing:
- **Real-time Transcription**: Convert speech to text using [Whisper](https://github.com/openai/whisper) (multi-language) or [Parakeet](https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2) (English) models
- **Speaker Diarization**: Identify and label different speakers using [Pyannote](https://github.com/pyannote/pyannote-audio) 3.1
- **Live Translation**: Translate audio content in real-time to many languages with [Facebook Seamless-M4T](https://github.com/facebookresearch/seamless_communication)
- **Topic Detection & Summarization**: Extract key topics and generate concise summaries using LLMs
- **Meeting Recording**: Create permanent records of meetings with searchable transcripts
Currently we provide [modal.com](https://modal.com/) gpu template to deploy.
## Background
The project architecture consists of three primary components:
- **Back-End**: Python server that offers an API and data persistence, found in `server/`.
- **Front-End**: NextJS React project hosted on Vercel, located in `www/`.
- **GPU implementation**: Providing services such as speech-to-text transcription, topic generation, automated summaries, and translations.
- **Back-End**: Python server that offers an API and data persistence, found in `server/`.
- **GPU implementation**: Providing services such as speech-to-text transcription, topic generation, automated summaries, and translations. Most reliable option is Modal deployment
It also uses authentik for authentication if activated.
It also uses authentik for authentication if activated, and Vercel for deployment and configuration of the front-end.
## Contribution Guidelines
@@ -89,8 +72,6 @@ Note: We currently do not have instructions for Windows users.
## Installation
*Note: we're working toward better installation, theses instructions are not accurate for now*
### Frontend
Start with `cd www`.
@@ -98,16 +79,17 @@ Start with `cd www`.
**Installation**
```bash
pnpm install
cp .env.example .env
yarn install
cp .env_template .env
cp config-template.ts config.ts
```
Then, fill in the environment variables in `.env` as needed. If you are unsure on how to proceed, ask in Zulip.
Then, fill in the environment variables in `.env` and the configuration in `config.ts` as needed. If you are unsure on how to proceed, ask in Zulip.
**Run in development mode**
```bash
pnpm dev
yarn dev
```
Then (after completing server setup and starting it) open [http://localhost:3000](http://localhost:3000) to view it in the browser.
@@ -117,7 +99,7 @@ Then (after completing server setup and starting it) open [http://localhost:3000
To generate the TypeScript files from the openapi.json file, make sure the python server is running, then run:
```bash
pnpm openapi
yarn openapi
```
### Backend
@@ -167,47 +149,3 @@ You can manually process an audio file by calling the process tool:
```bash
uv run python -m reflector.tools.process path/to/audio.wav
```
## Reprocessing any transcription
```bash
uv run -m reflector.tools.process_transcript 81ec38d1-9dd7-43d2-b3f8-51f4d34a07cd --sync
```
## Build-time env variables
Next.js projects are more used to NEXT_PUBLIC_ prefixed buildtime vars. We don't have those for the reason we need to serve a ccustomizable prebuild docker container.
Instead, all the variables are runtime. Variables needed to the frontend are served to the frontend app at initial render.
It also means there's no static prebuild and no static files to serve for js/html.
## Feature Flags
Reflector uses environment variable-based feature flags to control application functionality. These flags allow you to enable or disable features without code changes.
### Available Feature Flags
| Feature Flag | Environment Variable |
|-------------|---------------------|
| `requireLogin` | `FEATURE_REQUIRE_LOGIN` |
| `privacy` | `FEATURE_PRIVACY` |
| `browse` | `FEATURE_BROWSE` |
| `sendToZulip` | `FEATURE_SEND_TO_ZULIP` |
| `rooms` | `FEATURE_ROOMS` |
### Setting Feature Flags
Feature flags are controlled via environment variables using the pattern `FEATURE_{FEATURE_NAME}` where `{FEATURE_NAME}` is the SCREAMING_SNAKE_CASE version of the feature name.
**Examples:**
```bash
# Enable user authentication requirement
FEATURE_REQUIRE_LOGIN=true
# Disable browse functionality
FEATURE_BROWSE=false
# Enable Zulip integration
FEATURE_SEND_TO_ZULIP=true
```

586
REFACTOR_WHEREBY_FINDING.md Normal file
View File

@@ -0,0 +1,586 @@
# Whereby to Daily.co Migration Feasibility Analysis
## Executive Summary
After analysis of the current Whereby integration and Daily.co's capabilities, migrating to Daily.co is technically feasible. The migration can be done in phases:
1. **Phase 1**: Feature parity with current implementation (standard cloud recording)
2. **Phase 2**: Enhanced capabilities with raw-tracks recording for improved diarization
### Current Implementation Analysis
Based on code review:
- **Webhook handling**: The current webhook handler (`server/reflector/views/whereby.py`) only tracks `num_clients`, not individual participants
- **Focus management**: The frontend has 70+ lines managing focus between Whereby embed and consent dialog
- **Participant tracking**: No participant names or IDs are captured in the current implementation
- **Recording type**: Cloud recording to S3 in MP4 format with mixed audio
### Migration Approach
**Phase 1**: 1:1 feature replacement maintaining current functionality:
- Standard cloud recording (same as current Whereby implementation)
- Same recording workflow: Video platform → S3 → Reflector processing
- No changes to existing diarization or transcription pipeline
**Phase 2**: Enhanced capabilities (future implementation):
- Raw-tracks recording for speaker-separated audio
- Improved diarization with participant-to-audio mapping
- Per-participant transcription accuracy
## Current Whereby Integration Analysis
### Backend Integration
#### Core API Module (`server/reflector/whereby.py`)
- **Meeting Creation**: Creates rooms with S3 recording configuration
- **Session Monitoring**: Tracks meeting status via room sessions API
- **Logo Upload**: Handles branding for meetings
- **Key Functions**:
```python
create_meeting(room_name, logo_s3_url) -> dict
monitor_room_session(meeting_link) -> dict
upload_logo(file_stream, content_type) -> str
```
#### Webhook Handler (`server/reflector/views/whereby.py`)
- **Endpoint**: `/v1/whereby_webhook`
- **Security**: HMAC signature validation
- **Events Handled**:
- `room.participant.joined`
- `room.participant.left`
- **Pain Point**: Delay between actual join/leave and webhook delivery
#### Room Management (`server/reflector/views/rooms.py`)
- Creates meetings via Whereby API
- Stores meeting data in database
- Manages recording lifecycle
### Frontend Integration
#### Main Room Component (`www/app/[roomName]/page.tsx`)
- Uses `@whereby.com/browser-sdk` (v3.3.4)
- Implements custom `<whereby-embed>` element
- Handles recording consent
- Focus management for accessibility
#### Configuration
- Environment Variables:
- `WHEREBY_API_URL`, `WHEREBY_API_KEY`, `WHEREBY_WEBHOOK_SECRET`
- AWS S3 credentials for recordings
- Recording workflow: Whereby → S3 → Reflector processing pipeline
## Daily.co Capabilities Analysis
### REST API Features
#### Room Management
```
POST /rooms - Create room with configuration
GET /rooms/:name/presence - Real-time participant data
POST /rooms/:name/recordings/start - Start recording
```
#### Recording Options
```json
{
"enable_recording": "raw-tracks" // Key feature for diarization
}
```
#### Webhook Events
- `participant.joined` / `participant.left`
- `waiting-participant.joined` / `waiting-participant.left`
- `recording.started` / `recording.ready-to-download`
- `recording.error`
### React SDK (@daily-co/daily-react)
#### Modern Hook-based Architecture
```jsx
// Participant tracking
const participantIds = useParticipantIds({ filter: 'remote' });
const [username, videoState] = useParticipantProperty(id, ['user_name', 'tracks.video.state']);
// Recording management
const { isRecording, startRecording, stopRecording } = useRecording();
// Real-time participant data
const participants = useParticipants();
```
## Feature Comparison
| Feature | Whereby | Daily.co |
|---------|---------|----------|
| **Room Creation** | REST API | REST API |
| **Recording Types** | Cloud (MP4) | Cloud (MP4), Local, Raw-tracks |
| **S3 Integration** | Direct upload | Direct upload with IAM roles |
| **Frontend Integration** | Custom element | React hooks or iframe |
| **Webhooks** | HMAC verified | HMAC verified |
| **Participant Data** | Via webhooks | Via webhooks + Presence API |
| **Recording Trigger** | Automatic/manual | Automatic/manual |
## Migration Plan
### Phase 1: Backend API Client
#### 1.1 Create Daily.co API Client (`server/reflector/daily.py`)
```python
from datetime import datetime
import httpx
from reflector.db.rooms import Room
from reflector.settings import settings
class DailyClient:
def __init__(self):
self.base_url = "https://api.daily.co/v1"
self.headers = {
"Authorization": f"Bearer {settings.DAILY_API_KEY}",
"Content-Type": "application/json"
}
self.timeout = 10
async def create_meeting(self, room_name_prefix: str, end_date: datetime, room: Room) -> dict:
"""Create a Daily.co room matching current Whereby functionality."""
data = {
"name": f"{room_name_prefix}-{datetime.now().strftime('%Y%m%d%H%M%S')}",
"privacy": "private" if room.is_locked else "public",
"properties": {
"enable_recording": "raw-tracks", #"cloud",
"enable_chat": True,
"enable_screenshare": True,
"start_video_off": False,
"start_audio_off": False,
"exp": int(end_date.timestamp()),
"enable_recording_ui": False, # We handle consent ourselves
}
}
# if room.recording_type == "cloud":
data["properties"]["recording_bucket"] = {
"bucket_name": settings.AWS_S3_BUCKET,
"bucket_region": settings.AWS_REGION,
"assume_role_arn": settings.AWS_DAILY_ROLE_ARN,
"path": f"recordings/{data['name']}"
}
async with httpx.AsyncClient() as client:
response = await client.post(
f"{self.base_url}/rooms",
headers=self.headers,
json=data,
timeout=self.timeout
)
response.raise_for_status()
room_data = response.json()
# Return in Whereby-compatible format
return {
"roomUrl": room_data["url"],
"hostRoomUrl": room_data["url"] + "?t=" + room_data["config"]["token"],
"roomName": room_data["name"],
"meetingId": room_data["id"]
}
async def get_room_sessions(self, room_name: str) -> dict:
"""Get room session data (similar to Whereby's insights)."""
async with httpx.AsyncClient() as client:
response = await client.get(
f"{self.base_url}/rooms/{room_name}",
headers=self.headers,
timeout=self.timeout
)
response.raise_for_status()
return response.json()
```
#### 1.2 Update Webhook Handler (`server/reflector/views/daily.py`)
```python
import hmac
import json
from datetime import datetime
from hashlib import sha256
from fastapi import APIRouter, HTTPException, Request
from pydantic import BaseModel
from reflector.db.meetings import meetings_controller
from reflector.settings import settings
router = APIRouter()
class DailyWebhookEvent(BaseModel):
type: str
id: str
ts: int
data: dict
def verify_daily_webhook(body: bytes, signature: str) -> bool:
"""Verify Daily.co webhook signature."""
expected = hmac.new(
settings.DAILY_WEBHOOK_SECRET.encode(),
body,
sha256
).hexdigest()
return hmac.compare_digest(expected, signature)
@router.post("/daily")
async def daily_webhook(event: DailyWebhookEvent, request: Request):
# Verify webhook signature
body = await request.body()
signature = request.headers.get("X-Daily-Signature", "")
if not verify_daily_webhook(body, signature):
raise HTTPException(status_code=401, detail="Invalid webhook signature")
# Handle participant events
if event.type == "participant.joined":
meeting = await meetings_controller.get_by_room_name(event.data["room_name"])
if meeting:
# Update participant info immediately
await meetings_controller.add_participant(
meeting.id,
participant_id=event.data["participant"]["user_id"],
name=event.data["participant"]["user_name"],
joined_at=datetime.fromtimestamp(event.ts / 1000)
)
elif event.type == "participant.left":
meeting = await meetings_controller.get_by_room_name(event.data["room_name"])
if meeting:
await meetings_controller.remove_participant(
meeting.id,
participant_id=event.data["participant"]["user_id"],
left_at=datetime.fromtimestamp(event.ts / 1000)
)
elif event.type == "recording.ready-to-download":
# Process cloud recording (same as Whereby)
meeting = await meetings_controller.get_by_room_name(event.data["room_name"])
if meeting:
# Queue standard processing task
from reflector.worker.tasks import process_recording
process_recording.delay(
meeting_id=meeting.id,
recording_url=event.data["download_link"],
recording_id=event.data["recording_id"]
)
return {"status": "ok"}
```
### Phase 2: Frontend Components
#### 2.1 Replace Whereby SDK with Daily React
First, update dependencies:
```bash
# Remove Whereby
yarn remove @whereby.com/browser-sdk
# Add Daily.co
yarn add @daily-co/daily-react @daily-co/daily-js
```
#### 2.2 New Room Component (`www/app/[roomName]/page.tsx`)
```tsx
"use client";
import { useCallback, useEffect, useRef, useState } from "react";
import {
DailyProvider,
useDaily,
useParticipantIds,
useRecording,
useDailyEvent,
useLocalParticipant,
} from "@daily-co/daily-react";
import { Box, Button, Text, VStack, HStack, Spinner } from "@chakra-ui/react";
import { toaster } from "../components/ui/toaster";
import useRoomMeeting from "./useRoomMeeting";
import { useRouter } from "next/navigation";
import { notFound } from "next/navigation";
import useSessionStatus from "../lib/useSessionStatus";
import { useRecordingConsent } from "../recordingConsentContext";
import DailyIframe from "@daily-co/daily-js";
// Daily.co Call Interface Component
function CallInterface() {
const daily = useDaily();
const { isRecording, startRecording, stopRecording } = useRecording();
const localParticipant = useLocalParticipant();
const participantIds = useParticipantIds({ filter: "remote" });
// Real-time participant tracking
useDailyEvent("participant-joined", useCallback((event) => {
console.log(`${event.participant.user_name} joined the call`);
// No need for webhooks - we have immediate access!
}, []));
useDailyEvent("participant-left", useCallback((event) => {
console.log(`${event.participant.user_name} left the call`);
}, []));
return (
<Box position="relative" width="100vw" height="100vh">
{/* Daily.co automatically handles the video/audio UI */}
<Box
as="iframe"
src={daily?.iframe()?.src}
width="100%"
height="100%"
allow="camera; microphone; fullscreen; speaker; display-capture"
style={{ border: "none" }}
/>
{/* Recording status indicator */}
{isRecording && (
<Box
position="absolute"
top={4}
right={4}
bg="red.500"
color="white"
px={3}
py={1}
borderRadius="md"
fontSize="sm"
>
Recording
</Box>
)}
{/* Participant count with real-time data */}
<Box position="absolute" bottom={4} left={4} bg="gray.800" color="white" px={3} py={1} borderRadius="md">
Participants: {participantIds.length + 1}
</Box>
</Box>
);
}
// Main Room Component with Daily.co Integration
export default function Room({ params }: { params: { roomName: string } }) {
const roomName = params.roomName;
const meeting = useRoomMeeting(roomName);
const router = useRouter();
const { isLoading, isAuthenticated } = useSessionStatus();
const [dailyUrl, setDailyUrl] = useState<string | null>(null);
const [callFrame, setCallFrame] = useState<DailyIframe | null>(null);
// Initialize Daily.co call
useEffect(() => {
if (!meeting?.response?.room_url) return;
const frame = DailyIframe.createCallObject({
showLeaveButton: true,
showFullscreenButton: true,
});
frame.on("left-meeting", () => {
router.push("/browse");
});
setCallFrame(frame);
setDailyUrl(meeting.response.room_url);
return () => {
frame.destroy();
};
}, [meeting?.response?.room_url, router]);
if (isLoading) {
return (
<Box display="flex" justifyContent="center" alignItems="center" height="100vh">
<Spinner color="blue.500" size="xl" />
</Box>
);
}
if (!dailyUrl || !callFrame) {
return null;
}
return (
<DailyProvider callObject={callFrame} url={dailyUrl}>
<CallInterface />
<ConsentDialog meetingId={meeting?.response?.id} />
</DailyProvider>
);
}
### Phase 3: Testing & Validation
For Phase 1 (feature parity), the existing processing pipeline remains unchanged:
1. Daily.co records meeting to S3 (same as Whereby)
2. Webhook notifies when recording is ready
3. Existing pipeline downloads and processes the MP4 file
4. Current diarization and transcription tools continue to work
Key validation points:
- Recording format matches (MP4 with mixed audio)
- S3 paths are compatible
- Processing pipeline requires no changes
- Transcript quality remains the same
## Future Enhancement: Raw-Tracks Recording (Phase 2)
### Raw-Tracks Processing for Enhanced Diarization
Daily.co's raw-tracks recording provides individual audio streams per participant, enabling:
```python
@shared_task
def process_daily_raw_tracks(meeting_id: str, recording_id: str, tracks: list):
"""Process Daily.co raw-tracks with perfect speaker attribution."""
for track in tracks:
participant_id = track["participant_id"]
participant_name = track["participant_name"]
track_url = track["download_url"]
# Download individual participant audio
response = download_track(track_url)
# Process with known speaker identity
transcript = transcribe_audio(
audio_data=response.content,
speaker_id=participant_id,
speaker_name=participant_name
)
# Store with accurate speaker mapping
save_transcript_segment(
meeting_id=meeting_id,
speaker_id=participant_id,
text=transcript.text,
timestamps=transcript.timestamps
)
```
### Benefits of Raw-Tracks (Future)
1. **Deterministic Speaker Attribution**: Each audio track is already speaker-separated
2. **Improved Transcription Accuracy**: Clean audio without cross-talk
3. **Parallel Processing**: Process multiple speakers simultaneously
4. **Better Metrics**: Accurate talk-time per participant
### Phase 4: Database & Configuration
#### 4.1 Environment Variable Updates
Update `.env` files:
```bash
# Remove Whereby variables
# WHEREBY_API_URL=https://api.whereby.dev/v1
# WHEREBY_API_KEY=your-whereby-key
# WHEREBY_WEBHOOK_SECRET=your-whereby-secret
# AWS_WHEREBY_S3_BUCKET=whereby-recordings
# AWS_WHEREBY_ACCESS_KEY_ID=whereby-key
# AWS_WHEREBY_ACCESS_KEY_SECRET=whereby-secret
# Add Daily.co variables
DAILY_API_KEY=your-daily-api-key
DAILY_WEBHOOK_SECRET=your-daily-webhook-secret
AWS_DAILY_S3_BUCKET=daily-recordings
AWS_DAILY_ROLE_ARN=arn:aws:iam::123456789:role/daily-recording-role
AWS_REGION=us-west-2
```
#### 4.2 Database Migration
```sql
-- Alembic migration to support Daily.co
-- server/alembic/versions/xxx_migrate_to_daily.py
def upgrade():
# Add platform field to support gradual migration
op.add_column('rooms', sa.Column('platform', sa.String(), server_default='whereby'))
op.add_column('meetings', sa.Column('platform', sa.String(), server_default='whereby'))
# No other schema changes needed for feature parity
def downgrade():
op.drop_column('meetings', 'platform')
op.drop_column('rooms', 'platform')
```
#### 4.3 Settings Update (`server/reflector/settings.py`)
```python
class Settings(BaseSettings):
# Remove Whereby settings
# WHEREBY_API_URL: str = "https://api.whereby.dev/v1"
# WHEREBY_API_KEY: str
# WHEREBY_WEBHOOK_SECRET: str
# AWS_WHEREBY_S3_BUCKET: str
# AWS_WHEREBY_ACCESS_KEY_ID: str
# AWS_WHEREBY_ACCESS_KEY_SECRET: str
# Add Daily.co settings
DAILY_API_KEY: str
DAILY_WEBHOOK_SECRET: str
AWS_DAILY_S3_BUCKET: str
AWS_DAILY_ROLE_ARN: str
AWS_REGION: str = "us-west-2"
# Daily.co room URL pattern
DAILY_ROOM_URL_PATTERN: str = "https://{subdomain}.daily.co/{room_name}"
DAILY_SUBDOMAIN: str = "reflector" # Your Daily.co subdomain
```
## Technical Differences
### Phase 1 Implementation
1. **Frontend**: Replace `<whereby-embed>` custom element with Daily.co React components or iframe
2. **Backend**: Create Daily.co API client matching Whereby's functionality
3. **Webhooks**: Map Daily.co events to existing database operations
4. **Recording**: Maintain same MP4 format and S3 storage
### Phase 2 Capabilities (Future)
1. **Raw-tracks recording**: Individual audio streams per participant
2. **Presence API**: Real-time participant data without webhook delays
3. **Transcription API**: Built-in transcription services
4. **Advanced recording options**: Multiple formats and layouts
## Risks and Mitigation
### Risk 1: API Differences
- **Mitigation**: Create abstraction layer to minimize changes
- Comprehensive testing of all endpoints
### Risk 2: Recording Format Changes
- **Mitigation**: Build adapter for raw-tracks processing
- Maintain backward compatibility during transition
### Risk 3: User Experience Changes
- **Mitigation**: A/B testing with gradual rollout
- Feature parity checklist before full migration
## Recommendation
Migration to Daily.co is technically feasible and can be implemented in phases:
### Phase 1: Feature Parity
- Replace Whereby with Daily.co maintaining exact same functionality
- Use standard cloud recording (MP4 to S3)
- No changes to processing pipeline
### Phase 2: Enhanced Capabilities (Future)
- Enable raw-tracks recording for improved diarization
- Implement participant-level audio processing
- Add real-time features using Presence API
## Next Steps
1. Set up Daily.co account and obtain API credentials
2. Implement feature flag system for gradual migration
3. Create Daily.co API client matching Whereby functionality
4. Update frontend to support both platforms
5. Test thoroughly before rollout
---
*Analysis based on current codebase review and API documentation comparison.*

View File

@@ -6,7 +6,6 @@ services:
- 1250:1250
volumes:
- ./server/:/app/
- /app/.venv
env_file:
- ./server/.env
environment:
@@ -17,7 +16,6 @@ services:
context: server
volumes:
- ./server/:/app/
- /app/.venv
env_file:
- ./server/.env
environment:
@@ -28,7 +26,6 @@ services:
context: server
volumes:
- ./server/:/app/
- /app/.venv
env_file:
- ./server/.env
environment:
@@ -39,19 +36,16 @@ services:
ports:
- 6379:6379
web:
image: node:22-alpine
image: node:18
ports:
- "3000:3000"
command: sh -c "corepack enable && pnpm install && pnpm dev"
command: sh -c "yarn install && yarn dev"
restart: unless-stopped
working_dir: /app
volumes:
- ./www:/app/
- /app/node_modules
env_file:
- ./www/.env.local
environment:
- NODE_ENV=development
postgres:
image: postgres:17

View File

@@ -1,37 +0,0 @@
# Production Docker Compose configuration for Frontend
# Usage: docker compose -f docker-compose.prod.yml up -d
services:
web:
image: monadicalsas/reflector-frontend:latest
pull_policy: always
environment:
- KV_URL=${KV_URL:-redis://redis:6379}
- SITE_URL=${SITE_URL}
- API_URL=${API_URL}
- WEBSOCKET_URL=${WEBSOCKET_URL}
- NEXTAUTH_URL=${NEXTAUTH_URL:-http://localhost:3000}
- NEXTAUTH_SECRET=${NEXTAUTH_SECRET:-changeme-in-production}
- AUTHENTIK_ISSUER=${AUTHENTIK_ISSUER}
- AUTHENTIK_CLIENT_ID=${AUTHENTIK_CLIENT_ID}
- AUTHENTIK_CLIENT_SECRET=${AUTHENTIK_CLIENT_SECRET}
- AUTHENTIK_REFRESH_TOKEN_URL=${AUTHENTIK_REFRESH_TOKEN_URL}
- SENTRY_DSN=${SENTRY_DSN}
- SENTRY_IGNORE_API_RESOLUTION_ERROR=${SENTRY_IGNORE_API_RESOLUTION_ERROR:-1}
depends_on:
- redis
restart: unless-stopped
redis:
image: redis:7.2-alpine
restart: unless-stopped
healthcheck:
test: ["CMD", "redis-cli", "ping"]
interval: 30s
timeout: 3s
retries: 3
volumes:
- redis_data:/data
volumes:
redis_data:

View File

@@ -1,241 +0,0 @@
# Transcript Formats
The Reflector API provides multiple output formats for transcript data through the `transcript_format` query parameter on the GET `/v1/transcripts/{id}` endpoint.
## Overview
When retrieving a transcript, you can specify the desired format using the `transcript_format` query parameter. The API supports four formats optimized for different use cases:
- **text** - Plain text with speaker names (default)
- **text-timestamped** - Timestamped text with speaker names
- **webvtt-named** - WebVTT subtitle format with participant names
- **json** - Structured JSON segments with full metadata
All formats include participant information when available, resolving speaker IDs to actual names.
## Query Parameter Usage
```
GET /v1/transcripts/{id}?transcript_format={format}
```
### Parameters
- `transcript_format` (optional): The desired output format
- Type: `"text" | "text-timestamped" | "webvtt-named" | "json"`
- Default: `"text"`
## Format Descriptions
### Text Format (`text`)
**Use case:** Simple, human-readable transcript for display or export.
**Format:** Speaker names followed by their dialogue, one line per segment.
**Example:**
```
John Smith: Hello everyone
Jane Doe: Hi there
John Smith: How are you today?
```
**Request:**
```bash
GET /v1/transcripts/{id}?transcript_format=text
```
**Response:**
```json
{
"id": "transcript_123",
"name": "Meeting Recording",
"transcript_format": "text",
"transcript": "John Smith: Hello everyone\nJane Doe: Hi there\nJohn Smith: How are you today?",
"participants": [
{"id": "p1", "speaker": 0, "name": "John Smith"},
{"id": "p2", "speaker": 1, "name": "Jane Doe"}
],
...
}
```
### Text Timestamped Format (`text-timestamped`)
**Use case:** Transcript with timing information for navigation or reference.
**Format:** `[MM:SS]` timestamp prefix before each speaker and dialogue.
**Example:**
```
[00:00] John Smith: Hello everyone
[00:05] Jane Doe: Hi there
[00:12] John Smith: How are you today?
```
**Request:**
```bash
GET /v1/transcripts/{id}?transcript_format=text-timestamped
```
**Response:**
```json
{
"id": "transcript_123",
"name": "Meeting Recording",
"transcript_format": "text-timestamped",
"transcript": "[00:00] John Smith: Hello everyone\n[00:05] Jane Doe: Hi there\n[00:12] John Smith: How are you today?",
"participants": [
{"id": "p1", "speaker": 0, "name": "John Smith"},
{"id": "p2", "speaker": 1, "name": "Jane Doe"}
],
...
}
```
### WebVTT Named Format (`webvtt-named`)
**Use case:** Subtitle files for video players, accessibility tools, or video editing.
**Format:** Standard WebVTT subtitle format with voice tags using participant names.
**Example:**
```
WEBVTT
00:00:00.000 --> 00:00:05.000
<v John Smith>Hello everyone
00:00:05.000 --> 00:00:12.000
<v Jane Doe>Hi there
00:00:12.000 --> 00:00:18.000
<v John Smith>How are you today?
```
**Request:**
```bash
GET /v1/transcripts/{id}?transcript_format=webvtt-named
```
**Response:**
```json
{
"id": "transcript_123",
"name": "Meeting Recording",
"transcript_format": "webvtt-named",
"transcript": "WEBVTT\n\n00:00:00.000 --> 00:00:05.000\n<v John Smith>Hello everyone\n\n...",
"participants": [
{"id": "p1", "speaker": 0, "name": "John Smith"},
{"id": "p2", "speaker": 1, "name": "Jane Doe"}
],
...
}
```
### JSON Format (`json`)
**Use case:** Programmatic access with full timing and speaker metadata.
**Format:** Array of segment objects with speaker information, text content, and precise timing.
**Example:**
```json
[
{
"speaker": 0,
"speaker_name": "John Smith",
"text": "Hello everyone",
"start": 0.0,
"end": 5.0
},
{
"speaker": 1,
"speaker_name": "Jane Doe",
"text": "Hi there",
"start": 5.0,
"end": 12.0
},
{
"speaker": 0,
"speaker_name": "John Smith",
"text": "How are you today?",
"start": 12.0,
"end": 18.0
}
]
```
**Request:**
```bash
GET /v1/transcripts/{id}?transcript_format=json
```
**Response:**
```json
{
"id": "transcript_123",
"name": "Meeting Recording",
"transcript_format": "json",
"transcript": [
{
"speaker": 0,
"speaker_name": "John Smith",
"text": "Hello everyone",
"start": 0.0,
"end": 5.0
},
{
"speaker": 1,
"speaker_name": "Jane Doe",
"text": "Hi there",
"start": 5.0,
"end": 12.0
}
],
"participants": [
{"id": "p1", "speaker": 0, "name": "John Smith"},
{"id": "p2", "speaker": 1, "name": "Jane Doe"}
],
...
}
```
## Response Structure
All formats return the same base transcript metadata with an additional `transcript_format` field and format-specific `transcript` field:
### Common Fields
- `id`: Transcript identifier
- `user_id`: Owner user ID (if authenticated)
- `name`: Transcript name
- `status`: Processing status
- `locked`: Whether transcript is locked for editing
- `duration`: Total duration in seconds
- `title`: Auto-generated or custom title
- `short_summary`: Brief summary
- `long_summary`: Detailed summary
- `created_at`: Creation timestamp
- `share_mode`: Access control setting
- `source_language`: Original audio language
- `target_language`: Translation target language
- `reviewed`: Whether transcript has been reviewed
- `meeting_id`: Associated meeting ID (if applicable)
- `source_kind`: Source type (live, file, room)
- `room_id`: Associated room ID (if applicable)
- `audio_deleted`: Whether audio has been deleted
- `participants`: Array of participant objects with speaker mappings
### Format-Specific Fields
- `transcript_format`: The format identifier (discriminator field)
- `transcript`: The formatted transcript content (string for text/webvtt formats, array for json format)
## Speaker Name Resolution
All formats resolve speaker IDs to participant names when available:
- If a participant exists for the speaker ID, their name is used
- If no participant exists, a default name like "Speaker 0" is generated
- Speaker IDs are integers (0, 1, 2, etc.) assigned during diarization

View File

@@ -1,33 +0,0 @@
# OS / Editor
.DS_Store
.vscode/
.idea/
# Python
__pycache__/
*.py[cod]
*$py.class
# Logs
*.log
# Env and secrets
.env
.env.*
*.env
*.secret
# Build / dist
build/
dist/
.eggs/
*.egg-info/
# Coverage / test
.pytest_cache/
.coverage*
htmlcov/
# Modal local state (if any)
modal_mounts/
.modal_cache/

View File

@@ -1,171 +0,0 @@
# Reflector GPU implementation - Transcription and LLM
This repository hold an API for the GPU implementation of the Reflector API service,
and use [Modal.com](https://modal.com)
- `reflector_diarizer.py` - Diarization API
- `reflector_transcriber.py` - Transcription API (Whisper)
- `reflector_transcriber_parakeet.py` - Transcription API (NVIDIA Parakeet)
- `reflector_translator.py` - Translation API
## Modal.com deployment
Create a modal secret, and name it `reflector-gpu`.
It should contain an `REFLECTOR_APIKEY` environment variable with a value.
The deployment is done using [Modal.com](https://modal.com) service.
```
$ modal deploy reflector_transcriber.py
...
└── 🔨 Created web => https://xxxx--reflector-transcriber-web.modal.run
$ modal deploy reflector_transcriber_parakeet.py
...
└── 🔨 Created web => https://xxxx--reflector-transcriber-parakeet-web.modal.run
$ modal deploy reflector_llm.py
...
└── 🔨 Created web => https://xxxx--reflector-llm-web.modal.run
```
Then in your reflector api configuration `.env`, you can set these keys:
```
TRANSCRIPT_BACKEND=modal
TRANSCRIPT_URL=https://xxxx--reflector-transcriber-web.modal.run
TRANSCRIPT_MODAL_API_KEY=REFLECTOR_APIKEY
DIARIZATION_BACKEND=modal
DIARIZATION_URL=https://xxxx--reflector-diarizer-web.modal.run
DIARIZATION_MODAL_API_KEY=REFLECTOR_APIKEY
TRANSLATION_BACKEND=modal
TRANSLATION_URL=https://xxxx--reflector-translator-web.modal.run
TRANSLATION_MODAL_API_KEY=REFLECTOR_APIKEY
```
## API
Authentication must be passed with the `Authorization` header, using the `bearer` scheme.
```
Authorization: bearer <REFLECTOR_APIKEY>
```
### LLM
`POST /llm`
**request**
```
{
"prompt": "xxx"
}
```
**response**
```
{
"text": "xxx completed"
}
```
### Transcription
#### Parakeet Transcriber (`reflector_transcriber_parakeet.py`)
NVIDIA Parakeet is a state-of-the-art ASR model optimized for real-time transcription with superior word-level timestamps.
**GPU Configuration:**
- **A10G GPU** - Used for `/v1/audio/transcriptions` endpoint (small files, live transcription)
- Higher concurrency (max_inputs=10)
- Optimized for multiple small audio files
- Supports batch processing for efficiency
- **L40S GPU** - Used for `/v1/audio/transcriptions-from-url` endpoint (large files)
- Lower concurrency but more powerful processing
- Optimized for single large audio files
- VAD-based chunking for long-form audio
##### `/v1/audio/transcriptions` - Small file transcription
**request** (multipart/form-data)
- `file` or `files[]` - audio file(s) to transcribe
- `model` - model name (default: `nvidia/parakeet-tdt-0.6b-v2`)
- `language` - language code (default: `en`)
- `batch` - whether to use batch processing for multiple files (default: `true`)
**response**
```json
{
"text": "transcribed text",
"words": [
{"word": "hello", "start": 0.0, "end": 0.5},
{"word": "world", "start": 0.5, "end": 1.0}
],
"filename": "audio.mp3"
}
```
For multiple files with batch=true:
```json
{
"results": [
{
"filename": "audio1.mp3",
"text": "transcribed text",
"words": [...]
},
{
"filename": "audio2.mp3",
"text": "transcribed text",
"words": [...]
}
]
}
```
##### `/v1/audio/transcriptions-from-url` - Large file transcription
**request** (application/json)
```json
{
"audio_file_url": "https://example.com/audio.mp3",
"model": "nvidia/parakeet-tdt-0.6b-v2",
"language": "en",
"timestamp_offset": 0.0
}
```
**response**
```json
{
"text": "transcribed text from large file",
"words": [
{"word": "hello", "start": 0.0, "end": 0.5},
{"word": "world", "start": 0.5, "end": 1.0}
]
}
```
**Supported file types:** mp3, mp4, mpeg, mpga, m4a, wav, webm
#### Whisper Transcriber (`reflector_transcriber.py`)
`POST /transcribe`
**request** (multipart/form-data)
- `file` - audio file
- `language` - language code (e.g. `en`)
**response**
```
{
"text": "xxx",
"words": [
{"text": "xxx", "start": 0.0, "end": 1.0}
]
}
```

View File

@@ -1,253 +0,0 @@
"""
Reflector GPU backend - diarizer
===================================
"""
import os
import uuid
from typing import Mapping, NewType
from urllib.parse import urlparse
import modal
PYANNOTE_MODEL_NAME: str = "pyannote/speaker-diarization-3.1"
MODEL_DIR = "/root/diarization_models"
UPLOADS_PATH = "/uploads"
SUPPORTED_FILE_EXTENSIONS = ["mp3", "mp4", "mpeg", "mpga", "m4a", "wav", "webm"]
DiarizerUniqFilename = NewType("DiarizerUniqFilename", str)
AudioFileExtension = NewType("AudioFileExtension", str)
app = modal.App(name="reflector-diarizer")
# Volume for temporary file uploads
upload_volume = modal.Volume.from_name("diarizer-uploads", create_if_missing=True)
def detect_audio_format(url: str, headers: Mapping[str, str]) -> AudioFileExtension:
parsed_url = urlparse(url)
url_path = parsed_url.path
for ext in SUPPORTED_FILE_EXTENSIONS:
if url_path.lower().endswith(f".{ext}"):
return AudioFileExtension(ext)
content_type = headers.get("content-type", "").lower()
if "audio/mpeg" in content_type or "audio/mp3" in content_type:
return AudioFileExtension("mp3")
if "audio/wav" in content_type:
return AudioFileExtension("wav")
if "audio/mp4" in content_type:
return AudioFileExtension("mp4")
raise ValueError(
f"Unsupported audio format for URL: {url}. "
f"Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
)
def download_audio_to_volume(
audio_file_url: str,
) -> tuple[DiarizerUniqFilename, AudioFileExtension]:
import requests
from fastapi import HTTPException
print(f"Checking audio file at: {audio_file_url}")
response = requests.head(audio_file_url, allow_redirects=True)
if response.status_code == 404:
raise HTTPException(status_code=404, detail="Audio file not found")
print(f"Downloading audio file from: {audio_file_url}")
response = requests.get(audio_file_url, allow_redirects=True)
if response.status_code != 200:
print(f"Download failed with status {response.status_code}: {response.text}")
raise HTTPException(
status_code=response.status_code,
detail=f"Failed to download audio file: {response.status_code}",
)
audio_suffix = detect_audio_format(audio_file_url, response.headers)
unique_filename = DiarizerUniqFilename(f"{uuid.uuid4()}.{audio_suffix}")
file_path = f"{UPLOADS_PATH}/{unique_filename}"
print(f"Writing file to: {file_path} (size: {len(response.content)} bytes)")
with open(file_path, "wb") as f:
f.write(response.content)
upload_volume.commit()
print(f"File saved as: {unique_filename}")
return unique_filename, audio_suffix
def migrate_cache_llm():
"""
XXX The cache for model files in Transformers v4.22.0 has been updated.
Migrating your old cache. This is a one-time only operation. You can
interrupt this and resume the migration later on by calling
`transformers.utils.move_cache()`.
"""
from transformers.utils.hub import move_cache
print("Moving LLM cache")
move_cache(cache_dir=MODEL_DIR, new_cache_dir=MODEL_DIR)
print("LLM cache moved")
def download_pyannote_audio():
from pyannote.audio import Pipeline
Pipeline.from_pretrained(
PYANNOTE_MODEL_NAME,
cache_dir=MODEL_DIR,
use_auth_token=os.environ["HF_TOKEN"],
)
diarizer_image = (
modal.Image.debian_slim(python_version="3.10.8")
.pip_install(
"pyannote.audio==3.1.0",
"requests",
"onnx",
"torchaudio",
"onnxruntime-gpu",
"torch==2.0.0",
"transformers==4.34.0",
"sentencepiece",
"protobuf",
"numpy",
"huggingface_hub",
"hf-transfer",
)
.run_function(
download_pyannote_audio,
secrets=[modal.Secret.from_name("hf_token")],
)
.run_function(migrate_cache_llm)
.env(
{
"LD_LIBRARY_PATH": (
"/usr/local/lib/python3.10/site-packages/nvidia/cudnn/lib/:"
"/opt/conda/lib/python3.10/site-packages/nvidia/cublas/lib/"
)
}
)
)
@app.cls(
gpu="A100",
timeout=60 * 30,
image=diarizer_image,
volumes={UPLOADS_PATH: upload_volume},
enable_memory_snapshot=True,
experimental_options={"enable_gpu_snapshot": True},
secrets=[
modal.Secret.from_name("hf_token"),
],
)
@modal.concurrent(max_inputs=1)
class Diarizer:
@modal.enter(snap=True)
def enter(self):
import torch
from pyannote.audio import Pipeline
self.use_gpu = torch.cuda.is_available()
self.device = "cuda" if self.use_gpu else "cpu"
print(f"Using device: {self.device}")
self.diarization_pipeline = Pipeline.from_pretrained(
PYANNOTE_MODEL_NAME,
cache_dir=MODEL_DIR,
use_auth_token=os.environ["HF_TOKEN"],
)
self.diarization_pipeline.to(torch.device(self.device))
@modal.method()
def diarize(self, filename: str, timestamp: float = 0.0):
import torchaudio
upload_volume.reload()
file_path = f"{UPLOADS_PATH}/{filename}"
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
print(f"Diarizing audio from: {file_path}")
waveform, sample_rate = torchaudio.load(file_path)
diarization = self.diarization_pipeline(
{"waveform": waveform, "sample_rate": sample_rate}
)
words = []
for diarization_segment, _, speaker in diarization.itertracks(yield_label=True):
words.append(
{
"start": round(timestamp + diarization_segment.start, 3),
"end": round(timestamp + diarization_segment.end, 3),
"speaker": int(speaker[-2:]),
}
)
print("Diarization complete")
return {"diarization": words}
# -------------------------------------------------------------------
# Web API
# -------------------------------------------------------------------
@app.function(
timeout=60 * 10,
scaledown_window=60 * 3,
secrets=[
modal.Secret.from_name("reflector-gpu"),
],
volumes={UPLOADS_PATH: upload_volume},
image=diarizer_image,
)
@modal.concurrent(max_inputs=40)
@modal.asgi_app()
def web():
from fastapi import Depends, FastAPI, HTTPException, status
from fastapi.security import OAuth2PasswordBearer
from pydantic import BaseModel
diarizerstub = Diarizer()
app = FastAPI()
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
def apikey_auth(apikey: str = Depends(oauth2_scheme)):
if apikey != os.environ["REFLECTOR_GPU_APIKEY"]:
raise HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail="Invalid API key",
headers={"WWW-Authenticate": "Bearer"},
)
class DiarizationResponse(BaseModel):
result: dict
@app.post("/diarize", dependencies=[Depends(apikey_auth)])
def diarize(audio_file_url: str, timestamp: float = 0.0) -> DiarizationResponse:
unique_filename, audio_suffix = download_audio_to_volume(audio_file_url)
try:
func = diarizerstub.diarize.spawn(
filename=unique_filename, timestamp=timestamp
)
result = func.get()
return result
finally:
try:
file_path = f"{UPLOADS_PATH}/{unique_filename}"
print(f"Deleting file: {file_path}")
os.remove(file_path)
upload_volume.commit()
except Exception as e:
print(f"Error cleaning up {unique_filename}: {e}")
return app

View File

@@ -1,608 +0,0 @@
import os
import sys
import threading
import uuid
from typing import Generator, Mapping, NamedTuple, NewType, TypedDict
from urllib.parse import urlparse
import modal
MODEL_NAME = "large-v2"
MODEL_COMPUTE_TYPE: str = "float16"
MODEL_NUM_WORKERS: int = 1
MINUTES = 60 # seconds
SAMPLERATE = 16000
UPLOADS_PATH = "/uploads"
CACHE_PATH = "/models"
SUPPORTED_FILE_EXTENSIONS = ["mp3", "mp4", "mpeg", "mpga", "m4a", "wav", "webm"]
VAD_CONFIG = {
"batch_max_duration": 30.0,
"silence_padding": 0.5,
"window_size": 512,
}
WhisperUniqFilename = NewType("WhisperUniqFilename", str)
AudioFileExtension = NewType("AudioFileExtension", str)
app = modal.App("reflector-transcriber")
model_cache = modal.Volume.from_name("models", create_if_missing=True)
upload_volume = modal.Volume.from_name("whisper-uploads", create_if_missing=True)
class TimeSegment(NamedTuple):
"""Represents a time segment with start and end times."""
start: float
end: float
class AudioSegment(NamedTuple):
"""Represents an audio segment with timing and audio data."""
start: float
end: float
audio: any
class TranscriptResult(NamedTuple):
"""Represents a transcription result with text and word timings."""
text: str
words: list["WordTiming"]
class WordTiming(TypedDict):
"""Represents a word with its timing information."""
word: str
start: float
end: float
def download_model():
from faster_whisper import download_model
model_cache.reload()
download_model(MODEL_NAME, cache_dir=CACHE_PATH)
model_cache.commit()
image = (
modal.Image.debian_slim(python_version="3.12")
.env(
{
"HF_HUB_ENABLE_HF_TRANSFER": "1",
"LD_LIBRARY_PATH": (
"/usr/local/lib/python3.12/site-packages/nvidia/cudnn/lib/:"
"/opt/conda/lib/python3.12/site-packages/nvidia/cublas/lib/"
),
}
)
.apt_install("ffmpeg")
.pip_install(
"huggingface_hub==0.27.1",
"hf-transfer==0.1.9",
"torch==2.5.1",
"faster-whisper==1.1.1",
"fastapi==0.115.12",
"requests",
"librosa==0.10.1",
"numpy<2",
"silero-vad==5.1.0",
)
.run_function(download_model, volumes={CACHE_PATH: model_cache})
)
def detect_audio_format(url: str, headers: Mapping[str, str]) -> AudioFileExtension:
parsed_url = urlparse(url)
url_path = parsed_url.path
for ext in SUPPORTED_FILE_EXTENSIONS:
if url_path.lower().endswith(f".{ext}"):
return AudioFileExtension(ext)
content_type = headers.get("content-type", "").lower()
if "audio/mpeg" in content_type or "audio/mp3" in content_type:
return AudioFileExtension("mp3")
if "audio/wav" in content_type:
return AudioFileExtension("wav")
if "audio/mp4" in content_type:
return AudioFileExtension("mp4")
raise ValueError(
f"Unsupported audio format for URL: {url}. "
f"Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
)
def download_audio_to_volume(
audio_file_url: str,
) -> tuple[WhisperUniqFilename, AudioFileExtension]:
import requests
from fastapi import HTTPException
response = requests.head(audio_file_url, allow_redirects=True)
if response.status_code == 404:
raise HTTPException(status_code=404, detail="Audio file not found")
response = requests.get(audio_file_url, allow_redirects=True)
response.raise_for_status()
audio_suffix = detect_audio_format(audio_file_url, response.headers)
unique_filename = WhisperUniqFilename(f"{uuid.uuid4()}.{audio_suffix}")
file_path = f"{UPLOADS_PATH}/{unique_filename}"
with open(file_path, "wb") as f:
f.write(response.content)
upload_volume.commit()
return unique_filename, audio_suffix
def pad_audio(audio_array, sample_rate: int = SAMPLERATE):
"""Add 0.5s of silence if audio is shorter than the silence_padding window.
Whisper does not require this strictly, but aligning behavior with Parakeet
avoids edge-case crashes on extremely short inputs and makes comparisons easier.
"""
import numpy as np
audio_duration = len(audio_array) / sample_rate
if audio_duration < VAD_CONFIG["silence_padding"]:
silence_samples = int(sample_rate * VAD_CONFIG["silence_padding"])
silence = np.zeros(silence_samples, dtype=np.float32)
return np.concatenate([audio_array, silence])
return audio_array
@app.cls(
gpu="A10G",
timeout=5 * MINUTES,
scaledown_window=5 * MINUTES,
image=image,
volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
)
@modal.concurrent(max_inputs=10)
class TranscriberWhisperLive:
"""Live transcriber class for small audio segments (A10G).
Mirrors the Parakeet live class API but uses Faster-Whisper under the hood.
"""
@modal.enter()
def enter(self):
import faster_whisper
import torch
self.lock = threading.Lock()
self.use_gpu = torch.cuda.is_available()
self.device = "cuda" if self.use_gpu else "cpu"
self.model = faster_whisper.WhisperModel(
MODEL_NAME,
device=self.device,
compute_type=MODEL_COMPUTE_TYPE,
num_workers=MODEL_NUM_WORKERS,
download_root=CACHE_PATH,
local_files_only=True,
)
print(f"Model is on device: {self.device}")
@modal.method()
def transcribe_segment(
self,
filename: str,
language: str = "en",
):
"""Transcribe a single uploaded audio file by filename."""
upload_volume.reload()
file_path = f"{UPLOADS_PATH}/{filename}"
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
with self.lock:
with NoStdStreams():
segments, _ = self.model.transcribe(
file_path,
language=language,
beam_size=5,
word_timestamps=True,
vad_filter=True,
vad_parameters={"min_silence_duration_ms": 500},
)
segments = list(segments)
text = "".join(segment.text for segment in segments).strip()
words = [
{
"word": word.word,
"start": round(float(word.start), 2),
"end": round(float(word.end), 2),
}
for segment in segments
for word in segment.words
]
return {"text": text, "words": words}
@modal.method()
def transcribe_batch(
self,
filenames: list[str],
language: str = "en",
):
"""Transcribe multiple uploaded audio files and return per-file results."""
upload_volume.reload()
results = []
for filename in filenames:
file_path = f"{UPLOADS_PATH}/{filename}"
if not os.path.exists(file_path):
raise FileNotFoundError(f"Batch file not found: {file_path}")
with self.lock:
with NoStdStreams():
segments, _ = self.model.transcribe(
file_path,
language=language,
beam_size=5,
word_timestamps=True,
vad_filter=True,
vad_parameters={"min_silence_duration_ms": 500},
)
segments = list(segments)
text = "".join(seg.text for seg in segments).strip()
words = [
{
"word": w.word,
"start": round(float(w.start), 2),
"end": round(float(w.end), 2),
}
for seg in segments
for w in seg.words
]
results.append(
{
"filename": filename,
"text": text,
"words": words,
}
)
return results
@app.cls(
gpu="L40S",
timeout=15 * MINUTES,
image=image,
volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
)
class TranscriberWhisperFile:
"""File transcriber for larger/longer audio, using VAD-driven batching (L40S)."""
@modal.enter()
def enter(self):
import faster_whisper
import torch
from silero_vad import load_silero_vad
self.lock = threading.Lock()
self.use_gpu = torch.cuda.is_available()
self.device = "cuda" if self.use_gpu else "cpu"
self.model = faster_whisper.WhisperModel(
MODEL_NAME,
device=self.device,
compute_type=MODEL_COMPUTE_TYPE,
num_workers=MODEL_NUM_WORKERS,
download_root=CACHE_PATH,
local_files_only=True,
)
self.vad_model = load_silero_vad(onnx=False)
@modal.method()
def transcribe_segment(
self, filename: str, timestamp_offset: float = 0.0, language: str = "en"
):
import librosa
import numpy as np
from silero_vad import VADIterator
def vad_segments(
audio_array,
sample_rate: int = SAMPLERATE,
window_size: int = VAD_CONFIG["window_size"],
) -> Generator[TimeSegment, None, None]:
"""Generate speech segments as TimeSegment using Silero VAD."""
iterator = VADIterator(self.vad_model, sampling_rate=sample_rate)
start = None
for i in range(0, len(audio_array), window_size):
chunk = audio_array[i : i + window_size]
if len(chunk) < window_size:
chunk = np.pad(
chunk, (0, window_size - len(chunk)), mode="constant"
)
speech = iterator(chunk)
if not speech:
continue
if "start" in speech:
start = speech["start"]
continue
if "end" in speech and start is not None:
end = speech["end"]
yield TimeSegment(
start / float(SAMPLERATE), end / float(SAMPLERATE)
)
start = None
iterator.reset_states()
upload_volume.reload()
file_path = f"{UPLOADS_PATH}/{filename}"
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
audio_array, _sr = librosa.load(file_path, sr=SAMPLERATE, mono=True)
# Batch segments up to ~30s windows by merging contiguous VAD segments
merged_batches: list[TimeSegment] = []
batch_start = None
batch_end = None
max_duration = VAD_CONFIG["batch_max_duration"]
for segment in vad_segments(audio_array):
seg_start, seg_end = segment.start, segment.end
if batch_start is None:
batch_start, batch_end = seg_start, seg_end
continue
if seg_end - batch_start <= max_duration:
batch_end = seg_end
else:
merged_batches.append(TimeSegment(batch_start, batch_end))
batch_start, batch_end = seg_start, seg_end
if batch_start is not None and batch_end is not None:
merged_batches.append(TimeSegment(batch_start, batch_end))
all_text = []
all_words = []
for segment in merged_batches:
start_time, end_time = segment.start, segment.end
s_idx = int(start_time * SAMPLERATE)
e_idx = int(end_time * SAMPLERATE)
segment = audio_array[s_idx:e_idx]
segment = pad_audio(segment, SAMPLERATE)
with self.lock:
segments, _ = self.model.transcribe(
segment,
language=language,
beam_size=5,
word_timestamps=True,
vad_filter=True,
vad_parameters={"min_silence_duration_ms": 500},
)
segments = list(segments)
text = "".join(seg.text for seg in segments).strip()
words = [
{
"word": w.word,
"start": round(float(w.start) + start_time + timestamp_offset, 2),
"end": round(float(w.end) + start_time + timestamp_offset, 2),
}
for seg in segments
for w in seg.words
]
if text:
all_text.append(text)
all_words.extend(words)
return {"text": " ".join(all_text), "words": all_words}
def detect_audio_format(url: str, headers: dict) -> str:
from urllib.parse import urlparse
from fastapi import HTTPException
url_path = urlparse(url).path
for ext in SUPPORTED_FILE_EXTENSIONS:
if url_path.lower().endswith(f".{ext}"):
return ext
content_type = headers.get("content-type", "").lower()
if "audio/mpeg" in content_type or "audio/mp3" in content_type:
return "mp3"
if "audio/wav" in content_type:
return "wav"
if "audio/mp4" in content_type:
return "mp4"
raise HTTPException(
status_code=400,
detail=(
f"Unsupported audio format for URL. Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
),
)
def download_audio_to_volume(audio_file_url: str) -> tuple[str, str]:
import requests
from fastapi import HTTPException
response = requests.head(audio_file_url, allow_redirects=True)
if response.status_code == 404:
raise HTTPException(status_code=404, detail="Audio file not found")
response = requests.get(audio_file_url, allow_redirects=True)
response.raise_for_status()
audio_suffix = detect_audio_format(audio_file_url, response.headers)
unique_filename = f"{uuid.uuid4()}.{audio_suffix}"
file_path = f"{UPLOADS_PATH}/{unique_filename}"
with open(file_path, "wb") as f:
f.write(response.content)
upload_volume.commit()
return unique_filename, audio_suffix
@app.function(
scaledown_window=60,
timeout=600,
secrets=[
modal.Secret.from_name("reflector-gpu"),
],
volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
image=image,
)
@modal.concurrent(max_inputs=40)
@modal.asgi_app()
def web():
from fastapi import (
Body,
Depends,
FastAPI,
Form,
HTTPException,
UploadFile,
status,
)
from fastapi.security import OAuth2PasswordBearer
transcriber_live = TranscriberWhisperLive()
transcriber_file = TranscriberWhisperFile()
app = FastAPI()
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
def apikey_auth(apikey: str = Depends(oauth2_scheme)):
if apikey == os.environ["REFLECTOR_GPU_APIKEY"]:
return
raise HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail="Invalid API key",
headers={"WWW-Authenticate": "Bearer"},
)
class TranscriptResponse(dict):
pass
@app.post("/v1/audio/transcriptions", dependencies=[Depends(apikey_auth)])
def transcribe(
file: UploadFile = None,
files: list[UploadFile] | None = None,
model: str = Form(MODEL_NAME),
language: str = Form("en"),
batch: bool = Form(False),
):
if not file and not files:
raise HTTPException(
status_code=400, detail="Either 'file' or 'files' parameter is required"
)
if batch and not files:
raise HTTPException(
status_code=400, detail="Batch transcription requires 'files'"
)
upload_files = [file] if file else files
uploaded_filenames: list[str] = []
for upload_file in upload_files:
audio_suffix = upload_file.filename.split(".")[-1]
if audio_suffix not in SUPPORTED_FILE_EXTENSIONS:
raise HTTPException(
status_code=400,
detail=(
f"Unsupported audio format. Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
),
)
unique_filename = f"{uuid.uuid4()}.{audio_suffix}"
file_path = f"{UPLOADS_PATH}/{unique_filename}"
with open(file_path, "wb") as f:
content = upload_file.file.read()
f.write(content)
uploaded_filenames.append(unique_filename)
upload_volume.commit()
try:
if batch and len(upload_files) > 1:
func = transcriber_live.transcribe_batch.spawn(
filenames=uploaded_filenames,
language=language,
)
results = func.get()
return {"results": results}
results = []
for filename in uploaded_filenames:
func = transcriber_live.transcribe_segment.spawn(
filename=filename,
language=language,
)
result = func.get()
result["filename"] = filename
results.append(result)
return {"results": results} if len(results) > 1 else results[0]
finally:
for filename in uploaded_filenames:
try:
file_path = f"{UPLOADS_PATH}/{filename}"
os.remove(file_path)
except Exception:
pass
upload_volume.commit()
@app.post("/v1/audio/transcriptions-from-url", dependencies=[Depends(apikey_auth)])
def transcribe_from_url(
audio_file_url: str = Body(
..., description="URL of the audio file to transcribe"
),
model: str = Body(MODEL_NAME),
language: str = Body("en"),
timestamp_offset: float = Body(0.0),
):
unique_filename, _audio_suffix = download_audio_to_volume(audio_file_url)
try:
func = transcriber_file.transcribe_segment.spawn(
filename=unique_filename,
timestamp_offset=timestamp_offset,
language=language,
)
result = func.get()
return result
finally:
try:
file_path = f"{UPLOADS_PATH}/{unique_filename}"
os.remove(file_path)
upload_volume.commit()
except Exception:
pass
return app
class NoStdStreams:
def __init__(self):
self.devnull = open(os.devnull, "w")
def __enter__(self):
self._stdout, self._stderr = sys.stdout, sys.stderr
self._stdout.flush()
self._stderr.flush()
sys.stdout, sys.stderr = self.devnull, self.devnull
def __exit__(self, exc_type, exc_value, traceback):
sys.stdout, sys.stderr = self._stdout, self._stderr
self.devnull.close()

View File

@@ -1,663 +0,0 @@
import logging
import os
import sys
import threading
import uuid
from typing import Generator, Mapping, NamedTuple, NewType, TypedDict
from urllib.parse import urlparse
import modal
MODEL_NAME = "nvidia/parakeet-tdt-0.6b-v2"
SUPPORTED_FILE_EXTENSIONS = ["mp3", "mp4", "mpeg", "mpga", "m4a", "wav", "webm"]
SAMPLERATE = 16000
UPLOADS_PATH = "/uploads"
CACHE_PATH = "/cache"
VAD_CONFIG = {
"batch_max_duration": 30.0,
"silence_padding": 0.5,
"window_size": 512,
}
ParakeetUniqFilename = NewType("ParakeetUniqFilename", str)
AudioFileExtension = NewType("AudioFileExtension", str)
class TimeSegment(NamedTuple):
"""Represents a time segment with start and end times."""
start: float
end: float
class AudioSegment(NamedTuple):
"""Represents an audio segment with timing and audio data."""
start: float
end: float
audio: any
class TranscriptResult(NamedTuple):
"""Represents a transcription result with text and word timings."""
text: str
words: list["WordTiming"]
class WordTiming(TypedDict):
"""Represents a word with its timing information."""
word: str
start: float
end: float
app = modal.App("reflector-transcriber-parakeet")
# Volume for caching model weights
model_cache = modal.Volume.from_name("parakeet-model-cache", create_if_missing=True)
# Volume for temporary file uploads
upload_volume = modal.Volume.from_name("parakeet-uploads", create_if_missing=True)
image = (
modal.Image.from_registry(
"nvidia/cuda:12.8.0-cudnn-devel-ubuntu22.04", add_python="3.12"
)
.env(
{
"HF_HUB_ENABLE_HF_TRANSFER": "1",
"HF_HOME": "/cache",
"DEBIAN_FRONTEND": "noninteractive",
"CXX": "g++",
"CC": "g++",
}
)
.apt_install("ffmpeg")
.pip_install(
"hf_transfer==0.1.9",
"huggingface_hub[hf-xet]==0.31.2",
"nemo_toolkit[asr]==2.5.0",
"cuda-python==12.8.0",
"fastapi==0.115.12",
"numpy<2",
"librosa==0.11.0",
"requests",
"silero-vad==6.2.0",
"torch",
)
.entrypoint([]) # silence chatty logs by container on start
)
def detect_audio_format(url: str, headers: Mapping[str, str]) -> AudioFileExtension:
parsed_url = urlparse(url)
url_path = parsed_url.path
for ext in SUPPORTED_FILE_EXTENSIONS:
if url_path.lower().endswith(f".{ext}"):
return AudioFileExtension(ext)
content_type = headers.get("content-type", "").lower()
if "audio/mpeg" in content_type or "audio/mp3" in content_type:
return AudioFileExtension("mp3")
if "audio/wav" in content_type:
return AudioFileExtension("wav")
if "audio/mp4" in content_type:
return AudioFileExtension("mp4")
raise ValueError(
f"Unsupported audio format for URL: {url}. "
f"Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
)
def download_audio_to_volume(
audio_file_url: str,
) -> tuple[ParakeetUniqFilename, AudioFileExtension]:
import requests
from fastapi import HTTPException
response = requests.head(audio_file_url, allow_redirects=True)
if response.status_code == 404:
raise HTTPException(status_code=404, detail="Audio file not found")
response = requests.get(audio_file_url, allow_redirects=True)
response.raise_for_status()
audio_suffix = detect_audio_format(audio_file_url, response.headers)
unique_filename = ParakeetUniqFilename(f"{uuid.uuid4()}.{audio_suffix}")
file_path = f"{UPLOADS_PATH}/{unique_filename}"
with open(file_path, "wb") as f:
f.write(response.content)
upload_volume.commit()
return unique_filename, audio_suffix
def pad_audio(audio_array, sample_rate: int = SAMPLERATE):
"""Add 0.5 seconds of silence if audio is less than 500ms.
This is a workaround for a Parakeet bug where very short audio (<500ms) causes:
ValueError: `char_offsets`: [] and `processed_tokens`: [157, 834, 834, 841]
have to be of the same length
See: https://github.com/NVIDIA/NeMo/issues/8451
"""
import numpy as np
audio_duration = len(audio_array) / sample_rate
if audio_duration < 0.5:
silence_samples = int(sample_rate * 0.5)
silence = np.zeros(silence_samples, dtype=np.float32)
return np.concatenate([audio_array, silence])
return audio_array
@app.cls(
gpu="A10G",
timeout=600,
scaledown_window=300,
image=image,
volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
enable_memory_snapshot=True,
experimental_options={"enable_gpu_snapshot": True},
)
@modal.concurrent(max_inputs=10)
class TranscriberParakeetLive:
@modal.enter(snap=True)
def enter(self):
import nemo.collections.asr as nemo_asr
logging.getLogger("nemo_logger").setLevel(logging.CRITICAL)
self.lock = threading.Lock()
self.model = nemo_asr.models.ASRModel.from_pretrained(model_name=MODEL_NAME)
device = next(self.model.parameters()).device
print(f"Model is on device: {device}")
@modal.method()
def transcribe_segment(
self,
filename: str,
):
import librosa
upload_volume.reload()
file_path = f"{UPLOADS_PATH}/{filename}"
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
audio_array, sample_rate = librosa.load(file_path, sr=SAMPLERATE, mono=True)
padded_audio = pad_audio(audio_array, sample_rate)
with self.lock:
with NoStdStreams():
(output,) = self.model.transcribe([padded_audio], timestamps=True)
text = output.text.strip()
words: list[WordTiming] = [
WordTiming(
# XXX the space added here is to match the output of whisper
# whisper add space to each words, while parakeet don't
word=word_info["word"] + " ",
start=round(word_info["start"], 2),
end=round(word_info["end"], 2),
)
for word_info in output.timestamp["word"]
]
return {"text": text, "words": words}
@modal.method()
def transcribe_batch(
self,
filenames: list[str],
):
import librosa
upload_volume.reload()
results = []
audio_arrays = []
# Load all audio files with padding
for filename in filenames:
file_path = f"{UPLOADS_PATH}/{filename}"
if not os.path.exists(file_path):
raise FileNotFoundError(f"Batch file not found: {file_path}")
audio_array, sample_rate = librosa.load(file_path, sr=SAMPLERATE, mono=True)
padded_audio = pad_audio(audio_array, sample_rate)
audio_arrays.append(padded_audio)
with self.lock:
with NoStdStreams():
outputs = self.model.transcribe(audio_arrays, timestamps=True)
# Process results for each file
for i, (filename, output) in enumerate(zip(filenames, outputs)):
text = output.text.strip()
words: list[WordTiming] = [
WordTiming(
word=word_info["word"] + " ",
start=round(word_info["start"], 2),
end=round(word_info["end"], 2),
)
for word_info in output.timestamp["word"]
]
results.append(
{
"filename": filename,
"text": text,
"words": words,
}
)
return results
# L40S class for file transcription (bigger files)
@app.cls(
gpu="L40S",
timeout=900,
image=image,
volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
enable_memory_snapshot=True,
experimental_options={"enable_gpu_snapshot": True},
)
class TranscriberParakeetFile:
@modal.enter(snap=True)
def enter(self):
import nemo.collections.asr as nemo_asr
import torch
from silero_vad import load_silero_vad
logging.getLogger("nemo_logger").setLevel(logging.CRITICAL)
self.model = nemo_asr.models.ASRModel.from_pretrained(model_name=MODEL_NAME)
device = next(self.model.parameters()).device
print(f"Model is on device: {device}")
torch.set_num_threads(1)
self.vad_model = load_silero_vad(onnx=False)
print("Silero VAD initialized")
@modal.method()
def transcribe_segment(
self,
filename: str,
timestamp_offset: float = 0.0,
):
import librosa
import numpy as np
from silero_vad import VADIterator
def load_and_convert_audio(file_path):
audio_array, sample_rate = librosa.load(file_path, sr=SAMPLERATE, mono=True)
return audio_array
def vad_segment_generator(
audio_array,
) -> Generator[TimeSegment, None, None]:
"""Generate speech segments using VAD with start/end sample indices"""
vad_iterator = VADIterator(self.vad_model, sampling_rate=SAMPLERATE)
audio_duration = len(audio_array) / float(SAMPLERATE)
window_size = VAD_CONFIG["window_size"]
start = None
for i in range(0, len(audio_array), window_size):
chunk = audio_array[i : i + window_size]
if len(chunk) < window_size:
chunk = np.pad(
chunk, (0, window_size - len(chunk)), mode="constant"
)
speech_dict = vad_iterator(chunk)
if not speech_dict:
continue
if "start" in speech_dict:
start = speech_dict["start"]
continue
if "end" in speech_dict and start is not None:
end = speech_dict["end"]
start_time = start / float(SAMPLERATE)
end_time = end / float(SAMPLERATE)
yield TimeSegment(start_time, end_time)
start = None
if start is not None:
start_time = start / float(SAMPLERATE)
yield TimeSegment(start_time, audio_duration)
vad_iterator.reset_states()
def batch_speech_segments(
segments: Generator[TimeSegment, None, None], max_duration: int
) -> Generator[TimeSegment, None, None]:
"""
Input segments:
[0-2] [3-5] [6-8] [10-11] [12-15] [17-19] [20-22]
↓ (max_duration=10)
Output batches:
[0-8] [10-19] [20-22]
Note: silences are kept for better transcription, previous implementation was
passing segments separatly, but the output was less accurate.
"""
batch_start_time = None
batch_end_time = None
for segment in segments:
start_time, end_time = segment.start, segment.end
if batch_start_time is None or batch_end_time is None:
batch_start_time = start_time
batch_end_time = end_time
continue
total_duration = end_time - batch_start_time
if total_duration <= max_duration:
batch_end_time = end_time
continue
yield TimeSegment(batch_start_time, batch_end_time)
batch_start_time = start_time
batch_end_time = end_time
if batch_start_time is None or batch_end_time is None:
return
yield TimeSegment(batch_start_time, batch_end_time)
def batch_segment_to_audio_segment(
segments: Generator[TimeSegment, None, None],
audio_array,
) -> Generator[AudioSegment, None, None]:
"""Extract audio segments and apply padding for Parakeet compatibility.
Uses pad_audio to ensure segments are at least 0.5s long, preventing
Parakeet crashes. This padding may cause slight timing overlaps between
segments, which are corrected by enforce_word_timing_constraints.
"""
for segment in segments:
start_time, end_time = segment.start, segment.end
start_sample = int(start_time * SAMPLERATE)
end_sample = int(end_time * SAMPLERATE)
audio_segment = audio_array[start_sample:end_sample]
padded_segment = pad_audio(audio_segment, SAMPLERATE)
yield AudioSegment(start_time, end_time, padded_segment)
def transcribe_batch(model, audio_segments: list) -> list:
with NoStdStreams():
outputs = model.transcribe(audio_segments, timestamps=True)
return outputs
def enforce_word_timing_constraints(
words: list[WordTiming],
) -> list[WordTiming]:
"""Enforce that word end times don't exceed the start time of the next word.
Due to silence padding added in batch_segment_to_audio_segment for better
transcription accuracy, word timings from different segments may overlap.
This function ensures there are no overlaps by adjusting end times.
"""
if len(words) <= 1:
return words
enforced_words = []
for i, word in enumerate(words):
enforced_word = word.copy()
if i < len(words) - 1:
next_start = words[i + 1]["start"]
if enforced_word["end"] > next_start:
enforced_word["end"] = next_start
enforced_words.append(enforced_word)
return enforced_words
def emit_results(
results: list,
segments_info: list[AudioSegment],
) -> Generator[TranscriptResult, None, None]:
"""Yield transcribed text and word timings from model output, adjusting timestamps to absolute positions."""
for i, (output, segment) in enumerate(zip(results, segments_info)):
start_time, end_time = segment.start, segment.end
text = output.text.strip()
words: list[WordTiming] = [
WordTiming(
word=word_info["word"] + " ",
start=round(
word_info["start"] + start_time + timestamp_offset, 2
),
end=round(word_info["end"] + start_time + timestamp_offset, 2),
)
for word_info in output.timestamp["word"]
]
yield TranscriptResult(text, words)
upload_volume.reload()
file_path = f"{UPLOADS_PATH}/{filename}"
if not os.path.exists(file_path):
raise FileNotFoundError(f"File not found: {file_path}")
audio_array = load_and_convert_audio(file_path)
total_duration = len(audio_array) / float(SAMPLERATE)
all_text_parts: list[str] = []
all_words: list[WordTiming] = []
raw_segments = vad_segment_generator(audio_array)
speech_segments = batch_speech_segments(
raw_segments,
VAD_CONFIG["batch_max_duration"],
)
audio_segments = batch_segment_to_audio_segment(speech_segments, audio_array)
for batch in audio_segments:
audio_segment = batch.audio
results = transcribe_batch(self.model, [audio_segment])
for result in emit_results(
results,
[batch],
):
if not result.text:
continue
all_text_parts.append(result.text)
all_words.extend(result.words)
all_words = enforce_word_timing_constraints(all_words)
combined_text = " ".join(all_text_parts)
return {"text": combined_text, "words": all_words}
@app.function(
scaledown_window=60,
timeout=600,
secrets=[
modal.Secret.from_name("reflector-gpu"),
],
volumes={CACHE_PATH: model_cache, UPLOADS_PATH: upload_volume},
image=image,
)
@modal.concurrent(max_inputs=40)
@modal.asgi_app()
def web():
import os
import uuid
from fastapi import (
Body,
Depends,
FastAPI,
Form,
HTTPException,
UploadFile,
status,
)
from fastapi.security import OAuth2PasswordBearer
from pydantic import BaseModel
transcriber_live = TranscriberParakeetLive()
transcriber_file = TranscriberParakeetFile()
app = FastAPI()
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
def apikey_auth(apikey: str = Depends(oauth2_scheme)):
if apikey == os.environ["REFLECTOR_GPU_APIKEY"]:
return
raise HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail="Invalid API key",
headers={"WWW-Authenticate": "Bearer"},
)
class TranscriptResponse(BaseModel):
result: dict
@app.post("/v1/audio/transcriptions", dependencies=[Depends(apikey_auth)])
def transcribe(
file: UploadFile = None,
files: list[UploadFile] | None = None,
model: str = Form(MODEL_NAME),
language: str = Form("en"),
batch: bool = Form(False),
):
# Parakeet only supports English
if language != "en":
raise HTTPException(
status_code=400,
detail=f"Parakeet model only supports English. Got language='{language}'",
)
# Handle both single file and multiple files
if not file and not files:
raise HTTPException(
status_code=400, detail="Either 'file' or 'files' parameter is required"
)
if batch and not files:
raise HTTPException(
status_code=400, detail="Batch transcription requires 'files'"
)
upload_files = [file] if file else files
# Upload files to volume
uploaded_filenames = []
for upload_file in upload_files:
audio_suffix = upload_file.filename.split(".")[-1]
assert audio_suffix in SUPPORTED_FILE_EXTENSIONS
# Generate unique filename
unique_filename = f"{uuid.uuid4()}.{audio_suffix}"
file_path = f"{UPLOADS_PATH}/{unique_filename}"
print(f"Writing file to: {file_path}")
with open(file_path, "wb") as f:
content = upload_file.file.read()
f.write(content)
uploaded_filenames.append(unique_filename)
upload_volume.commit()
try:
# Use A10G live transcriber for per-file transcription
if batch and len(upload_files) > 1:
# Use batch transcription
func = transcriber_live.transcribe_batch.spawn(
filenames=uploaded_filenames,
)
results = func.get()
return {"results": results}
# Per-file transcription
results = []
for filename in uploaded_filenames:
func = transcriber_live.transcribe_segment.spawn(
filename=filename,
)
result = func.get()
result["filename"] = filename
results.append(result)
return {"results": results} if len(results) > 1 else results[0]
finally:
for filename in uploaded_filenames:
try:
file_path = f"{UPLOADS_PATH}/{filename}"
print(f"Deleting file: {file_path}")
os.remove(file_path)
except Exception as e:
print(f"Error deleting {filename}: {e}")
upload_volume.commit()
@app.post("/v1/audio/transcriptions-from-url", dependencies=[Depends(apikey_auth)])
def transcribe_from_url(
audio_file_url: str = Body(
..., description="URL of the audio file to transcribe"
),
model: str = Body(MODEL_NAME),
language: str = Body("en", description="Language code (only 'en' supported)"),
timestamp_offset: float = Body(0.0),
):
# Parakeet only supports English
if language != "en":
raise HTTPException(
status_code=400,
detail=f"Parakeet model only supports English. Got language='{language}'",
)
unique_filename, audio_suffix = download_audio_to_volume(audio_file_url)
try:
func = transcriber_file.transcribe_segment.spawn(
filename=unique_filename,
timestamp_offset=timestamp_offset,
)
result = func.get()
return result
finally:
try:
file_path = f"{UPLOADS_PATH}/{unique_filename}"
print(f"Deleting file: {file_path}")
os.remove(file_path)
upload_volume.commit()
except Exception as e:
print(f"Error cleaning up {unique_filename}: {e}")
return app
class NoStdStreams:
def __init__(self):
self.devnull = open(os.devnull, "w")
def __enter__(self):
self._stdout, self._stderr = sys.stdout, sys.stderr
self._stdout.flush()
self._stderr.flush()
sys.stdout, sys.stderr = self.devnull, self.devnull
def __exit__(self, exc_type, exc_value, traceback):
sys.stdout, sys.stderr = self._stdout, self._stderr
self.devnull.close()

View File

@@ -1,2 +0,0 @@
REFLECTOR_GPU_APIKEY=
HF_TOKEN=

View File

@@ -1,38 +0,0 @@
cache/
# OS / Editor
.DS_Store
.vscode/
.idea/
# Python
__pycache__/
*.py[cod]
*$py.class
# Env and secrets
.env
*.env
*.secret
HF_TOKEN
REFLECTOR_GPU_APIKEY
# Virtual env / uv
.venv/
venv/
ENV/
uv/
# Build / dist
build/
dist/
.eggs/
*.egg-info/
# Coverage / test
.pytest_cache/
.coverage*
htmlcov/
# Logs
*.log

View File

@@ -1,46 +0,0 @@
FROM python:3.12-slim
ENV PYTHONUNBUFFERED=1 \
UV_LINK_MODE=copy \
UV_NO_CACHE=1
WORKDIR /tmp
RUN apt-get update \
&& apt-get install -y \
ffmpeg \
curl \
ca-certificates \
gnupg \
wget \
&& apt-get clean
# Add NVIDIA CUDA repo for Debian 12 (bookworm) and install cuDNN 9 for CUDA 12
ADD https://developer.download.nvidia.com/compute/cuda/repos/debian12/x86_64/cuda-keyring_1.1-1_all.deb /cuda-keyring.deb
RUN dpkg -i /cuda-keyring.deb \
&& rm /cuda-keyring.deb \
&& apt-get update \
&& apt-get install -y --no-install-recommends \
cuda-cudart-12-6 \
libcublas-12-6 \
libcudnn9-cuda-12 \
libcudnn9-dev-cuda-12 \
&& apt-get clean \
&& rm -rf /var/lib/apt/lists/*
ADD https://astral.sh/uv/install.sh /uv-installer.sh
RUN sh /uv-installer.sh && rm /uv-installer.sh
ENV PATH="/root/.local/bin/:$PATH"
ENV LD_LIBRARY_PATH="/usr/local/cuda/lib64:/usr/lib/x86_64-linux-gnu:$LD_LIBRARY_PATH"
RUN mkdir -p /app
WORKDIR /app
COPY pyproject.toml uv.lock /app/
COPY ./app /app/app
COPY ./main.py /app/
COPY ./runserver.sh /app/
EXPOSE 8000
CMD ["sh", "/app/runserver.sh"]

View File

@@ -1,73 +0,0 @@
# Self-hosted Model API
Run transcription, translation, and diarization services compatible with Reflector's GPU Model API. Works on CPU or GPU.
Environment variables
- REFLECTOR_GPU_APIKEY: Optional Bearer token. If unset, auth is disabled.
- HF_TOKEN: Optional. Required for diarization to download pyannote pipelines
Requirements
- FFmpeg must be installed and on PATH (used for URL-based and segmented transcription)
- Python 3.12+
- NVIDIA GPU optional. If available, it will be used automatically
Local run
Set env vars in self_hosted/.env file
uv sync
uv run uvicorn main:app --host 0.0.0.0 --port 8000
Authentication
- If REFLECTOR_GPU_APIKEY is set, include header: Authorization: Bearer <key>
Endpoints
- POST /v1/audio/transcriptions
- multipart/form-data
- fields: file (single file) OR files[] (multiple files), language, batch (true/false)
- response: single { text, words, filename } or { results: [ ... ] }
- POST /v1/audio/transcriptions-from-url
- application/json
- body: { audio_file_url, language, timestamp_offset }
- response: { text, words }
- POST /translate
- text: query parameter
- body (application/json): { source_language, target_language }
- response: { text: { <src>: original, <tgt>: translated } }
- POST /diarize
- query parameters: audio_file_url, timestamp (optional)
- requires HF_TOKEN to be set (for pyannote)
- response: { diarization: [ { start, end, speaker } ] }
OpenAPI docs
- Visit /docs when the server is running
Docker
- Not yet provided in this directory. A Dockerfile will be added later. For now, use Local run above
Conformance tests
# From this directory
TRANSCRIPT_URL=http://localhost:8000 \
TRANSCRIPT_API_KEY=dev-key \
uv run -m pytest -m model_api --no-cov ../../server/tests/test_model_api_transcript.py
TRANSLATION_URL=http://localhost:8000 \
TRANSLATION_API_KEY=dev-key \
uv run -m pytest -m model_api --no-cov ../../server/tests/test_model_api_translation.py
DIARIZATION_URL=http://localhost:8000 \
DIARIZATION_API_KEY=dev-key \
uv run -m pytest -m model_api --no-cov ../../server/tests/test_model_api_diarization.py

View File

@@ -1,19 +0,0 @@
import os
from fastapi import Depends, HTTPException, status
from fastapi.security import OAuth2PasswordBearer
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
def apikey_auth(apikey: str = Depends(oauth2_scheme)):
required_key = os.environ.get("REFLECTOR_GPU_APIKEY")
if not required_key:
return
if apikey == required_key:
return
raise HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail="Invalid API key",
headers={"WWW-Authenticate": "Bearer"},
)

View File

@@ -1,12 +0,0 @@
from pathlib import Path
SUPPORTED_FILE_EXTENSIONS = ["mp3", "mp4", "mpeg", "mpga", "m4a", "wav", "webm"]
SAMPLE_RATE = 16000
VAD_CONFIG = {
"batch_max_duration": 30.0,
"silence_padding": 0.5,
"window_size": 512,
}
# App-level paths
UPLOADS_PATH = Path("/tmp/whisper-uploads")

View File

@@ -1,30 +0,0 @@
from contextlib import asynccontextmanager
from fastapi import FastAPI
from .routers.diarization import router as diarization_router
from .routers.transcription import router as transcription_router
from .routers.translation import router as translation_router
from .services.transcriber import WhisperService
from .services.diarizer import PyannoteDiarizationService
from .utils import ensure_dirs
@asynccontextmanager
async def lifespan(app: FastAPI):
ensure_dirs()
whisper_service = WhisperService()
whisper_service.load()
app.state.whisper = whisper_service
diarization_service = PyannoteDiarizationService()
diarization_service.load()
app.state.diarizer = diarization_service
yield
def create_app() -> FastAPI:
app = FastAPI(lifespan=lifespan)
app.include_router(transcription_router)
app.include_router(translation_router)
app.include_router(diarization_router)
return app

View File

@@ -1,30 +0,0 @@
from typing import List
from fastapi import APIRouter, Depends, Request
from pydantic import BaseModel
from ..auth import apikey_auth
from ..services.diarizer import PyannoteDiarizationService
from ..utils import download_audio_file
router = APIRouter(tags=["diarization"])
class DiarizationSegment(BaseModel):
start: float
end: float
speaker: int
class DiarizationResponse(BaseModel):
diarization: List[DiarizationSegment]
@router.post(
"/diarize", dependencies=[Depends(apikey_auth)], response_model=DiarizationResponse
)
def diarize(request: Request, audio_file_url: str, timestamp: float = 0.0):
with download_audio_file(audio_file_url) as (file_path, _ext):
file_path = str(file_path)
diarizer: PyannoteDiarizationService = request.app.state.diarizer
return diarizer.diarize_file(file_path, timestamp=timestamp)

View File

@@ -1,109 +0,0 @@
import uuid
from typing import Optional, Union
from fastapi import APIRouter, Body, Depends, Form, HTTPException, Request, UploadFile
from pydantic import BaseModel
from pathlib import Path
from ..auth import apikey_auth
from ..config import SUPPORTED_FILE_EXTENSIONS, UPLOADS_PATH
from ..services.transcriber import MODEL_NAME
from ..utils import cleanup_uploaded_files, download_audio_file
router = APIRouter(prefix="/v1/audio", tags=["transcription"])
class WordTiming(BaseModel):
word: str
start: float
end: float
class TranscriptResult(BaseModel):
text: str
words: list[WordTiming]
filename: Optional[str] = None
class TranscriptBatchResponse(BaseModel):
results: list[TranscriptResult]
@router.post(
"/transcriptions",
dependencies=[Depends(apikey_auth)],
response_model=Union[TranscriptResult, TranscriptBatchResponse],
)
def transcribe(
request: Request,
file: UploadFile = None,
files: list[UploadFile] | None = None,
model: str = Form(MODEL_NAME),
language: str = Form("en"),
batch: bool = Form(False),
):
service = request.app.state.whisper
if not file and not files:
raise HTTPException(
status_code=400, detail="Either 'file' or 'files' parameter is required"
)
if batch and not files:
raise HTTPException(
status_code=400, detail="Batch transcription requires 'files'"
)
upload_files = [file] if file else files
uploaded_paths: list[Path] = []
with cleanup_uploaded_files(uploaded_paths):
for upload_file in upload_files:
audio_suffix = upload_file.filename.split(".")[-1].lower()
if audio_suffix not in SUPPORTED_FILE_EXTENSIONS:
raise HTTPException(
status_code=400,
detail=(
f"Unsupported audio format. Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
),
)
unique_filename = f"{uuid.uuid4()}.{audio_suffix}"
file_path = UPLOADS_PATH / unique_filename
with open(file_path, "wb") as f:
content = upload_file.file.read()
f.write(content)
uploaded_paths.append(file_path)
if batch and len(upload_files) > 1:
results = []
for path in uploaded_paths:
result = service.transcribe_file(str(path), language=language)
result["filename"] = path.name
results.append(result)
return {"results": results}
results = []
for path in uploaded_paths:
result = service.transcribe_file(str(path), language=language)
result["filename"] = path.name
results.append(result)
return {"results": results} if len(results) > 1 else results[0]
@router.post(
"/transcriptions-from-url",
dependencies=[Depends(apikey_auth)],
response_model=TranscriptResult,
)
def transcribe_from_url(
request: Request,
audio_file_url: str = Body(..., description="URL of the audio file to transcribe"),
model: str = Body(MODEL_NAME),
language: str = Body("en"),
timestamp_offset: float = Body(0.0),
):
service = request.app.state.whisper
with download_audio_file(audio_file_url) as (file_path, _ext):
file_path = str(file_path)
result = service.transcribe_vad_url_segment(
file_path=file_path, timestamp_offset=timestamp_offset, language=language
)
return result

View File

@@ -1,28 +0,0 @@
from typing import Dict
from fastapi import APIRouter, Body, Depends
from pydantic import BaseModel
from ..auth import apikey_auth
from ..services.translator import TextTranslatorService
router = APIRouter(tags=["translation"])
translator = TextTranslatorService()
class TranslationResponse(BaseModel):
text: Dict[str, str]
@router.post(
"/translate",
dependencies=[Depends(apikey_auth)],
response_model=TranslationResponse,
)
def translate(
text: str,
source_language: str = Body("en"),
target_language: str = Body("fr"),
):
return translator.translate(text, source_language, target_language)

View File

@@ -1,42 +0,0 @@
import os
import threading
import torch
import torchaudio
from pyannote.audio import Pipeline
class PyannoteDiarizationService:
def __init__(self):
self._pipeline = None
self._device = "cpu"
self._lock = threading.Lock()
def load(self):
self._device = "cuda" if torch.cuda.is_available() else "cpu"
self._pipeline = Pipeline.from_pretrained(
"pyannote/speaker-diarization-3.1",
use_auth_token=os.environ.get("HF_TOKEN"),
)
self._pipeline.to(torch.device(self._device))
def diarize_file(self, file_path: str, timestamp: float = 0.0) -> dict:
if self._pipeline is None:
self.load()
waveform, sample_rate = torchaudio.load(file_path)
with self._lock:
diarization = self._pipeline(
{"waveform": waveform, "sample_rate": sample_rate}
)
words = []
for diarization_segment, _, speaker in diarization.itertracks(yield_label=True):
words.append(
{
"start": round(timestamp + diarization_segment.start, 3),
"end": round(timestamp + diarization_segment.end, 3),
"speaker": int(speaker[-2:])
if speaker and speaker[-2:].isdigit()
else 0,
}
)
return {"diarization": words}

View File

@@ -1,208 +0,0 @@
import os
import shutil
import subprocess
import threading
from typing import Generator
import faster_whisper
import librosa
import numpy as np
import torch
from fastapi import HTTPException
from silero_vad import VADIterator, load_silero_vad
from ..config import SAMPLE_RATE, VAD_CONFIG
# Whisper configuration (service-local defaults)
MODEL_NAME = "large-v2"
# None delegates compute type to runtime: float16 on CUDA, int8 on CPU
MODEL_COMPUTE_TYPE = None
MODEL_NUM_WORKERS = 1
CACHE_PATH = os.path.join(os.path.expanduser("~"), ".cache", "reflector-whisper")
from ..utils import NoStdStreams
class WhisperService:
def __init__(self):
self.model = None
self.device = "cpu"
self.lock = threading.Lock()
def load(self):
self.device = "cuda" if torch.cuda.is_available() else "cpu"
compute_type = MODEL_COMPUTE_TYPE or (
"float16" if self.device == "cuda" else "int8"
)
self.model = faster_whisper.WhisperModel(
MODEL_NAME,
device=self.device,
compute_type=compute_type,
num_workers=MODEL_NUM_WORKERS,
download_root=CACHE_PATH,
)
def pad_audio(self, audio_array, sample_rate: int = SAMPLE_RATE):
audio_duration = len(audio_array) / sample_rate
if audio_duration < VAD_CONFIG["silence_padding"]:
silence_samples = int(sample_rate * VAD_CONFIG["silence_padding"])
silence = np.zeros(silence_samples, dtype=np.float32)
return np.concatenate([audio_array, silence])
return audio_array
def enforce_word_timing_constraints(self, words: list[dict]) -> list[dict]:
if len(words) <= 1:
return words
enforced: list[dict] = []
for i, word in enumerate(words):
current = dict(word)
if i < len(words) - 1:
next_start = words[i + 1]["start"]
if current["end"] > next_start:
current["end"] = next_start
enforced.append(current)
return enforced
def transcribe_file(self, file_path: str, language: str = "en") -> dict:
input_for_model: str | "object" = file_path
try:
audio_array, _sample_rate = librosa.load(
file_path, sr=SAMPLE_RATE, mono=True
)
if len(audio_array) / float(SAMPLE_RATE) < VAD_CONFIG["silence_padding"]:
input_for_model = self.pad_audio(audio_array, SAMPLE_RATE)
except Exception:
pass
with self.lock:
with NoStdStreams():
segments, _ = self.model.transcribe(
input_for_model,
language=language,
beam_size=5,
word_timestamps=True,
vad_filter=True,
vad_parameters={"min_silence_duration_ms": 500},
)
segments = list(segments)
text = "".join(segment.text for segment in segments).strip()
words = [
{
"word": word.word,
"start": round(float(word.start), 2),
"end": round(float(word.end), 2),
}
for segment in segments
for word in segment.words
]
words = self.enforce_word_timing_constraints(words)
return {"text": text, "words": words}
def transcribe_vad_url_segment(
self, file_path: str, timestamp_offset: float = 0.0, language: str = "en"
) -> dict:
def load_audio_via_ffmpeg(input_path: str, sample_rate: int) -> np.ndarray:
ffmpeg_bin = shutil.which("ffmpeg") or "ffmpeg"
cmd = [
ffmpeg_bin,
"-nostdin",
"-threads",
"1",
"-i",
input_path,
"-f",
"f32le",
"-acodec",
"pcm_f32le",
"-ac",
"1",
"-ar",
str(sample_rate),
"pipe:1",
]
try:
proc = subprocess.run(
cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=True
)
except Exception as e:
raise HTTPException(status_code=400, detail=f"ffmpeg failed: {e}")
audio = np.frombuffer(proc.stdout, dtype=np.float32)
return audio
def vad_segments(
audio_array,
sample_rate: int = SAMPLE_RATE,
window_size: int = VAD_CONFIG["window_size"],
) -> Generator[tuple[float, float], None, None]:
vad_model = load_silero_vad(onnx=False)
iterator = VADIterator(vad_model, sampling_rate=sample_rate)
start = None
for i in range(0, len(audio_array), window_size):
chunk = audio_array[i : i + window_size]
if len(chunk) < window_size:
chunk = np.pad(
chunk, (0, window_size - len(chunk)), mode="constant"
)
speech = iterator(chunk)
if not speech:
continue
if "start" in speech:
start = speech["start"]
continue
if "end" in speech and start is not None:
end = speech["end"]
yield (start / float(SAMPLE_RATE), end / float(SAMPLE_RATE))
start = None
iterator.reset_states()
audio_array = load_audio_via_ffmpeg(file_path, SAMPLE_RATE)
merged_batches: list[tuple[float, float]] = []
batch_start = None
batch_end = None
max_duration = VAD_CONFIG["batch_max_duration"]
for seg_start, seg_end in vad_segments(audio_array):
if batch_start is None:
batch_start, batch_end = seg_start, seg_end
continue
if seg_end - batch_start <= max_duration:
batch_end = seg_end
else:
merged_batches.append((batch_start, batch_end))
batch_start, batch_end = seg_start, seg_end
if batch_start is not None and batch_end is not None:
merged_batches.append((batch_start, batch_end))
all_text = []
all_words = []
for start_time, end_time in merged_batches:
s_idx = int(start_time * SAMPLE_RATE)
e_idx = int(end_time * SAMPLE_RATE)
segment = audio_array[s_idx:e_idx]
segment = self.pad_audio(segment, SAMPLE_RATE)
with self.lock:
segments, _ = self.model.transcribe(
segment,
language=language,
beam_size=5,
word_timestamps=True,
vad_filter=True,
vad_parameters={"min_silence_duration_ms": 500},
)
segments = list(segments)
text = "".join(seg.text for seg in segments).strip()
words = [
{
"word": w.word,
"start": round(float(w.start) + start_time + timestamp_offset, 2),
"end": round(float(w.end) + start_time + timestamp_offset, 2),
}
for seg in segments
for w in seg.words
]
if text:
all_text.append(text)
all_words.extend(words)
all_words = self.enforce_word_timing_constraints(all_words)
return {"text": " ".join(all_text), "words": all_words}

View File

@@ -1,44 +0,0 @@
import threading
from transformers import MarianMTModel, MarianTokenizer, pipeline
class TextTranslatorService:
"""Simple text-to-text translator using HuggingFace MarianMT models.
This mirrors the modal translator API shape but uses text translation only.
"""
def __init__(self):
self._pipeline = None
self._lock = threading.Lock()
def load(self, source_language: str = "en", target_language: str = "fr"):
# Pick a default MarianMT model pair if available; fall back to Helsinki-NLP en->fr
model_name = self._resolve_model_name(source_language, target_language)
tokenizer = MarianTokenizer.from_pretrained(model_name)
model = MarianMTModel.from_pretrained(model_name)
self._pipeline = pipeline("translation", model=model, tokenizer=tokenizer)
def _resolve_model_name(self, src: str, tgt: str) -> str:
# Minimal mapping; extend as needed
pair = (src.lower(), tgt.lower())
mapping = {
("en", "fr"): "Helsinki-NLP/opus-mt-en-fr",
("fr", "en"): "Helsinki-NLP/opus-mt-fr-en",
("en", "es"): "Helsinki-NLP/opus-mt-en-es",
("es", "en"): "Helsinki-NLP/opus-mt-es-en",
("en", "de"): "Helsinki-NLP/opus-mt-en-de",
("de", "en"): "Helsinki-NLP/opus-mt-de-en",
}
return mapping.get(pair, "Helsinki-NLP/opus-mt-en-fr")
def translate(self, text: str, source_language: str, target_language: str) -> dict:
if self._pipeline is None:
self.load(source_language, target_language)
with self._lock:
results = self._pipeline(
text, src_lang=source_language, tgt_lang=target_language
)
translated = results[0]["translation_text"] if results else ""
return {"text": {source_language: text, target_language: translated}}

View File

@@ -1,107 +0,0 @@
import logging
import os
import sys
import uuid
from contextlib import contextmanager
from typing import Mapping
from urllib.parse import urlparse
from pathlib import Path
import requests
from fastapi import HTTPException
from .config import SUPPORTED_FILE_EXTENSIONS, UPLOADS_PATH
logger = logging.getLogger(__name__)
class NoStdStreams:
def __init__(self):
self.devnull = open(os.devnull, "w")
def __enter__(self):
self._stdout, self._stderr = sys.stdout, sys.stderr
self._stdout.flush()
self._stderr.flush()
sys.stdout, sys.stderr = self.devnull, self.devnull
def __exit__(self, exc_type, exc_value, traceback):
sys.stdout, sys.stderr = self._stdout, self._stderr
self.devnull.close()
def ensure_dirs():
UPLOADS_PATH.mkdir(parents=True, exist_ok=True)
def detect_audio_format(url: str, headers: Mapping[str, str]) -> str:
url_path = urlparse(url).path
for ext in SUPPORTED_FILE_EXTENSIONS:
if url_path.lower().endswith(f".{ext}"):
return ext
content_type = headers.get("content-type", "").lower()
if "audio/mpeg" in content_type or "audio/mp3" in content_type:
return "mp3"
if "audio/wav" in content_type:
return "wav"
if "audio/mp4" in content_type:
return "mp4"
raise HTTPException(
status_code=400,
detail=(
f"Unsupported audio format for URL. Supported extensions: {', '.join(SUPPORTED_FILE_EXTENSIONS)}"
),
)
def download_audio_to_uploads(audio_file_url: str) -> tuple[Path, str]:
response = requests.head(audio_file_url, allow_redirects=True)
if response.status_code == 404:
raise HTTPException(status_code=404, detail="Audio file not found")
response = requests.get(audio_file_url, allow_redirects=True)
response.raise_for_status()
audio_suffix = detect_audio_format(audio_file_url, response.headers)
unique_filename = f"{uuid.uuid4()}.{audio_suffix}"
file_path: Path = UPLOADS_PATH / unique_filename
with open(file_path, "wb") as f:
f.write(response.content)
return file_path, audio_suffix
@contextmanager
def download_audio_file(audio_file_url: str):
"""Download an audio file to UPLOADS_PATH and remove it after use.
Yields (file_path: Path, audio_suffix: str).
"""
file_path, audio_suffix = download_audio_to_uploads(audio_file_url)
try:
yield file_path, audio_suffix
finally:
try:
file_path.unlink(missing_ok=True)
except Exception as e:
logger.error("Error deleting temporary file %s: %s", file_path, e)
@contextmanager
def cleanup_uploaded_files(file_paths: list[Path]):
"""Ensure provided file paths are removed after use.
The provided list can be populated inside the context; all present entries
at exit will be deleted.
"""
try:
yield file_paths
finally:
for path in list(file_paths):
try:
path.unlink(missing_ok=True)
except Exception as e:
logger.error("Error deleting temporary file %s: %s", path, e)

View File

@@ -1,10 +0,0 @@
services:
reflector_gpu:
build:
context: .
ports:
- "8000:8000"
env_file:
- .env
volumes:
- ./cache:/root/.cache

View File

@@ -1,3 +0,0 @@
from app.factory import create_app
app = create_app()

View File

@@ -1,19 +0,0 @@
[project]
name = "reflector-gpu"
version = "0.1.0"
description = "Self-hosted GPU service for speech transcription, diarization, and translation via FastAPI."
readme = "README.md"
requires-python = ">=3.12"
dependencies = [
"fastapi[standard]>=0.116.1",
"uvicorn[standard]>=0.30.0",
"torch>=2.3.0",
"faster-whisper>=1.1.0",
"librosa==0.10.1",
"numpy<2",
"silero-vad==5.1.0",
"transformers>=4.35.0",
"sentencepiece",
"pyannote.audio==3.1.0",
"torchaudio>=2.3.0",
]

View File

@@ -1,17 +0,0 @@
#!/bin/sh
set -e
export PATH="/root/.local/bin:$PATH"
cd /app
# Install Python dependencies at runtime (first run or when FORCE_SYNC=1)
if [ ! -d "/app/.venv" ] || [ "$FORCE_SYNC" = "1" ]; then
echo "[startup] Installing Python dependencies with uv..."
uv sync --compile-bytecode --locked
else
echo "[startup] Using existing virtual environment at /app/.venv"
fi
exec uv run uvicorn main:app --host 0.0.0.0 --port 8000

3013
gpu/self_hosted/uv.lock generated

File diff suppressed because it is too large Load Diff

3
server/.gitignore vendored
View File

@@ -176,8 +176,7 @@ artefacts/
audio_*.wav
# ignore local database
*.sqlite3
*.db
reflector.sqlite3
data/
dump.rdb

View File

@@ -1,8 +1,7 @@
FROM python:3.12-slim
ENV PYTHONUNBUFFERED=1 \
UV_LINK_MODE=copy \
UV_NO_CACHE=1
UV_LINK_MODE=copy
# builder install base dependencies
WORKDIR /tmp
@@ -14,8 +13,8 @@ ENV PATH="/root/.local/bin/:$PATH"
# install application dependencies
RUN mkdir -p /app
WORKDIR /app
COPY pyproject.toml uv.lock README.md /app/
RUN uv sync --compile-bytecode --locked
COPY pyproject.toml uv.lock /app/
RUN touch README.md && env uv sync --compile-bytecode --locked
# pre-download nltk packages
RUN uv run python -c "import nltk; nltk.download('punkt_tab'); nltk.download('averaged_perceptron_tagger_eng')"
@@ -27,15 +26,4 @@ COPY migrations /app/migrations
COPY reflector /app/reflector
WORKDIR /app
# Create symlink for libgomp if it doesn't exist (for ARM64 compatibility)
RUN if [ "$(uname -m)" = "aarch64" ] && [ ! -f /usr/lib/libgomp.so.1 ]; then \
LIBGOMP_PATH=$(find /app/.venv/lib -path "*/torch.libs/libgomp*.so.*" 2>/dev/null | head -n1); \
if [ -n "$LIBGOMP_PATH" ]; then \
ln -sf "$LIBGOMP_PATH" /usr/lib/libgomp.so.1; \
fi \
fi
# Pre-check just to make sure the image will not fail
RUN uv run python -c "import silero_vad.model"
CMD ["./runserver.sh"]

View File

@@ -1,29 +1,3 @@
## API Key Management
### Finding Your User ID
```bash
# Get your OAuth sub (user ID) - requires authentication
curl -H "Authorization: Bearer <your_jwt>" http://localhost:1250/v1/me
# Returns: {"sub": "your-oauth-sub-here", "email": "...", ...}
```
### Creating API Keys
```bash
curl -X POST http://localhost:1250/v1/user/api-keys \
-H "Authorization: Bearer <your_jwt>" \
-H "Content-Type: application/json" \
-d '{"name": "My API Key"}'
```
### Using API Keys
```bash
# Use X-API-Key header instead of Authorization
curl -H "X-API-Key: <your_api_key>" http://localhost:1250/v1/transcripts
```
## AWS S3/SQS usage clarification
Whereby.com uploads recordings directly to our S3 bucket when meetings end.
@@ -66,5 +40,3 @@ uv run python -c "from reflector.pipelines.main_live_pipeline import task_pipeli
```bash
uv run python -c "from reflector.pipelines.main_live_pipeline import pipeline_post; pipeline_post(transcript_id='TRANSCRIPT_ID')"
```
.

View File

@@ -1,95 +0,0 @@
# Data Retention and Cleanup
## Overview
For public instances of Reflector, a data retention policy is automatically enforced to delete anonymous user data after a configurable period (default: 7 days). This ensures compliance with privacy expectations and prevents unbounded storage growth.
## Configuration
### Environment Variables
- `PUBLIC_MODE` (bool): Must be set to `true` to enable automatic cleanup
- `PUBLIC_DATA_RETENTION_DAYS` (int): Number of days to retain anonymous data (default: 7)
### What Gets Deleted
When data reaches the retention period, the following items are automatically removed:
1. **Transcripts** from anonymous users (where `user_id` is NULL):
- Database records
- Local files (audio.wav, audio.mp3, audio.json waveform)
- Storage files (cloud storage if configured)
## Automatic Cleanup
### Celery Beat Schedule
When `PUBLIC_MODE=true`, a Celery beat task runs daily at 3 AM to clean up old data:
```python
# Automatically scheduled when PUBLIC_MODE=true
"cleanup_old_public_data": {
"task": "reflector.worker.cleanup.cleanup_old_public_data",
"schedule": crontab(hour=3, minute=0), # Daily at 3 AM
}
```
### Running the Worker
Ensure both Celery worker and beat scheduler are running:
```bash
# Start Celery worker
uv run celery -A reflector.worker.app worker --loglevel=info
# Start Celery beat scheduler (in another terminal)
uv run celery -A reflector.worker.app beat
```
## Manual Cleanup
For testing or manual intervention, use the cleanup tool:
```bash
# Delete data older than 7 days (default)
uv run python -m reflector.tools.cleanup_old_data
# Delete data older than 30 days
uv run python -m reflector.tools.cleanup_old_data --days 30
```
Note: The manual tool uses the same implementation as the Celery worker task to ensure consistency.
## Important Notes
1. **User Data Deletion**: Only anonymous data (where `user_id` is NULL) is deleted. Authenticated user data is preserved.
2. **Storage Cleanup**: The system properly cleans up both local files and cloud storage when configured.
3. **Error Handling**: If individual deletions fail, the cleanup continues and logs errors. Failed deletions are reported in the task output.
4. **Public Instance Only**: The automatic cleanup task only runs when `PUBLIC_MODE=true` to prevent accidental data loss in private deployments.
## Testing
Run the cleanup tests:
```bash
uv run pytest tests/test_cleanup.py -v
```
## Monitoring
Check Celery logs for cleanup task execution:
```bash
# Look for cleanup task logs
grep "cleanup_old_public_data" celery.log
grep "Starting cleanup of old public data" celery.log
```
Task statistics are logged after each run:
- Number of transcripts deleted
- Number of meetings deleted
- Number of orphaned recordings deleted
- Any errors encountered

View File

@@ -1,194 +0,0 @@
## Reflector GPU Transcription API (Specification)
This document defines the Reflector GPU transcription API that all implementations must adhere to. Current implementations include NVIDIA Parakeet (NeMo) and Whisper (faster-whisper), both deployed on Modal.com. The API surface and response shapes are OpenAI/Whisper-compatible, so clients can switch implementations by changing only the base URL.
### Base URL and Authentication
- Example base URLs (Modal web endpoints):
- Parakeet: `https://<account>--reflector-transcriber-parakeet-web.modal.run`
- Whisper: `https://<account>--reflector-transcriber-web.modal.run`
- All endpoints are served under `/v1` and require a Bearer token:
```
Authorization: Bearer <REFLECTOR_GPU_APIKEY>
```
Note: To switch implementations, deploy the desired variant and point `TRANSCRIPT_URL` to its base URL. The API is identical.
### Supported file types
`mp3, mp4, mpeg, mpga, m4a, wav, webm`
### Models and languages
- Parakeet (NVIDIA NeMo): default `nvidia/parakeet-tdt-0.6b-v2`
- Language support: only `en`. Other languages return HTTP 400.
- Whisper (faster-whisper): default `large-v2` (or deployment-specific)
- Language support: multilingual (per Whisper model capabilities).
Note: The `model` parameter is accepted by all implementations for interface parity. Some backends may treat it as informational.
### Endpoints
#### POST /v1/audio/transcriptions
Transcribe one or more uploaded audio files.
Request: multipart/form-data
- `file` (File) — optional. Single file to transcribe.
- `files` (File[]) — optional. One or more files to transcribe.
- `model` (string) — optional. Defaults to the implementation-specific model (see above).
- `language` (string) — optional, defaults to `en`.
- Parakeet: only `en` is accepted; other values return HTTP 400
- Whisper: model-dependent; typically multilingual
- `batch` (boolean) — optional, defaults to `false`.
Notes:
- Provide either `file` or `files`, not both. If neither is provided, HTTP 400.
- `batch` requires `files`; using `batch=true` without `files` returns HTTP 400.
- Response shape for multiple files is the same regardless of `batch`.
- Files sent to this endpoint are processed in a single pass (no VAD/chunking). This is intended for short clips (roughly ≤ 30s; depends on GPU memory/model). For longer audio, prefer `/v1/audio/transcriptions-from-url` which supports VAD-based chunking.
Responses
Single file response:
```json
{
"text": "transcribed text",
"words": [
{ "word": "hello", "start": 0.0, "end": 0.5 },
{ "word": "world", "start": 0.5, "end": 1.0 }
],
"filename": "audio.mp3"
}
```
Multiple files response:
```json
{
"results": [
{"filename": "a1.mp3", "text": "...", "words": [...]},
{"filename": "a2.mp3", "text": "...", "words": [...]}]
}
```
Notes:
- Word objects always include keys: `word`, `start`, `end`.
- Some implementations may include a trailing space in `word` to match Whisper tokenization behavior; clients should trim if needed.
Example curl (single file):
```bash
curl -X POST \
-H "Authorization: Bearer $REFLECTOR_GPU_APIKEY" \
-F "file=@/path/to/audio.mp3" \
-F "language=en" \
"$BASE_URL/v1/audio/transcriptions"
```
Example curl (multiple files, batch):
```bash
curl -X POST \
-H "Authorization: Bearer $REFLECTOR_GPU_APIKEY" \
-F "files=@/path/a1.mp3" -F "files=@/path/a2.mp3" \
-F "batch=true" -F "language=en" \
"$BASE_URL/v1/audio/transcriptions"
```
#### POST /v1/audio/transcriptions-from-url
Transcribe a single remote audio file by URL.
Request: application/json
Body parameters:
- `audio_file_url` (string) — required. URL of the audio file to transcribe.
- `model` (string) — optional. Defaults to the implementation-specific model (see above).
- `language` (string) — optional, defaults to `en`. Parakeet only accepts `en`.
- `timestamp_offset` (number) — optional, defaults to `0.0`. Added to each word's `start`/`end` in the response.
```json
{
"audio_file_url": "https://example.com/audio.mp3",
"model": "nvidia/parakeet-tdt-0.6b-v2",
"language": "en",
"timestamp_offset": 0.0
}
```
Response:
```json
{
"text": "transcribed text",
"words": [
{ "word": "hello", "start": 10.0, "end": 10.5 },
{ "word": "world", "start": 10.5, "end": 11.0 }
]
}
```
Notes:
- `timestamp_offset` is added to each words `start`/`end` in the response.
- Implementations may perform VAD-based chunking and batching for long-form audio; word timings are adjusted accordingly.
Example curl:
```bash
curl -X POST \
-H "Authorization: Bearer $REFLECTOR_GPU_APIKEY" \
-H "Content-Type: application/json" \
-d '{
"audio_file_url": "https://example.com/audio.mp3",
"language": "en",
"timestamp_offset": 0
}' \
"$BASE_URL/v1/audio/transcriptions-from-url"
```
### Error handling
- 400 Bad Request
- Parakeet: `language` other than `en`
- Missing required parameters (`file`/`files` for upload; `audio_file_url` for URL endpoint)
- Unsupported file extension
- 401 Unauthorized
- Missing or invalid Bearer token
- 404 Not Found
- `audio_file_url` does not exist
### Implementation details
- GPUs: A10G for small-file/live, L40S for large-file URL transcription (subject to deployment)
- VAD chunking and segment batching; word timings adjusted and overlapping ends constrained
- Pads very short segments (< 0.5s) to avoid model crashes on some backends
### Server configuration (Reflector API)
Set the Reflector server to use the Modal backend and point `TRANSCRIPT_URL` to your chosen deployment:
```
TRANSCRIPT_BACKEND=modal
TRANSCRIPT_URL=https://<account>--reflector-transcriber-parakeet-web.modal.run
TRANSCRIPT_MODAL_API_KEY=<REFLECTOR_GPU_APIKEY>
```
### Conformance tests
Use the pytest-based conformance tests to validate any new implementation (including self-hosted) against this spec:
```
TRANSCRIPT_URL=https://<your-deployment-base> \
TRANSCRIPT_MODAL_API_KEY=your-api-key \
uv run -m pytest -m model_api --no-cov server/tests/test_model_api_transcript.py
```

View File

@@ -1,236 +0,0 @@
# Reflector Architecture: Whereby + Daily.co Recording Storage
## System Overview
```mermaid
graph TB
subgraph "Actors"
APP[Our App<br/>Reflector]
WHEREBY[Whereby Service<br/>External]
DAILY[Daily.co Service<br/>External]
end
subgraph "AWS S3 Buckets"
TRANSCRIPT_BUCKET[Transcript Bucket<br/>reflector-transcripts<br/>Output: Processed MP3s]
WHEREBY_BUCKET[Whereby Bucket<br/>reflector-whereby-recordings<br/>Input: Raw MP4s]
DAILY_BUCKET[Daily.co Bucket<br/>reflector-dailyco-recordings<br/>Input: Raw WebM tracks]
end
subgraph "AWS Infrastructure"
SQS[SQS Queue<br/>Whereby notifications]
end
subgraph "Database"
DB[(PostgreSQL<br/>Recordings, Transcripts, Meetings)]
end
APP -->|Write processed| TRANSCRIPT_BUCKET
APP -->|Read/Delete| WHEREBY_BUCKET
APP -->|Read/Delete| DAILY_BUCKET
APP -->|Poll| SQS
APP -->|Store metadata| DB
WHEREBY -->|Write recordings| WHEREBY_BUCKET
WHEREBY_BUCKET -->|S3 Event| SQS
WHEREBY -->|Participant webhooks<br/>room.client.joined/left| APP
DAILY -->|Write recordings| DAILY_BUCKET
DAILY -->|Recording webhook<br/>recording.ready-to-download| APP
```
**Note on Webhook vs S3 Event for Recording Processing:**
- **Whereby**: Uses S3 Events → SQS for recording availability (S3 as source of truth, no race conditions)
- **Daily.co**: Uses webhooks for recording availability (more immediate, built-in reliability)
- **Both**: Use webhooks for participant tracking (real-time updates)
## Credentials & Permissions
```mermaid
graph LR
subgraph "Master Credentials"
MASTER[TRANSCRIPT_STORAGE_AWS_*<br/>Access Key ID + Secret]
end
subgraph "Whereby Upload Credentials"
WHEREBY_CREDS[AWS_WHEREBY_ACCESS_KEY_*<br/>Access Key ID + Secret]
end
subgraph "Daily.co Upload Role"
DAILY_ROLE[DAILY_STORAGE_AWS_ROLE_ARN<br/>IAM Role ARN]
end
subgraph "Our App Uses"
MASTER -->|Read/Write/Delete| TRANSCRIPT_BUCKET[Transcript Bucket]
MASTER -->|Read/Delete| WHEREBY_BUCKET[Whereby Bucket]
MASTER -->|Read/Delete| DAILY_BUCKET[Daily.co Bucket]
MASTER -->|Poll/Delete| SQS[SQS Queue]
end
subgraph "We Give To Services"
WHEREBY_CREDS -->|Passed in API call| WHEREBY_SERVICE[Whereby Service]
WHEREBY_SERVICE -->|Write Only| WHEREBY_BUCKET
DAILY_ROLE -->|Passed in API call| DAILY_SERVICE[Daily.co Service]
DAILY_SERVICE -->|Assume Role| DAILY_ROLE
DAILY_SERVICE -->|Write Only| DAILY_BUCKET
end
```
# Video Platform Recording Integration
This document explains how Reflector receives and identifies multitrack audio recordings from different video platforms.
## Platform Comparison
| Platform | Delivery Method | Track Identification |
|----------|----------------|---------------------|
| **Daily.co** | Webhook | Explicit track list in payload |
| **Whereby** | SQS (S3 notifications) | Single file per notification |
---
## Daily.co
**Note:** Primary discovery via polling (`poll_daily_recordings`), webhooks as backup.
Daily.co uses **webhooks** to notify Reflector when recordings are ready.
### How It Works
1. **Daily.co sends webhook** when recording is ready
- Event type: `recording.ready-to-download`
- Endpoint: `/v1/daily/webhook` (`reflector/views/daily.py:46-102`)
2. **Webhook payload explicitly includes track list**:
```json
{
"recording_id": "7443ee0a-dab1-40eb-b316-33d6c0d5ff88",
"room_name": "daily-20251020193458",
"tracks": [
{
"type": "audio",
"s3Key": "monadical/daily-20251020193458/1760988935484-52f7f48b-fbab-431f-9a50-87b9abfc8255-cam-audio-1760988935922",
"size": 831843
},
{
"type": "audio",
"s3Key": "monadical/daily-20251020193458/1760988935484-a37c35e3-6f8e-4274-a482-e9d0f102a732-cam-audio-1760988943823",
"size": 408438
},
{
"type": "video",
"s3Key": "monadical/daily-20251020193458/...-video.webm",
"size": 30000000
}
]
}
```
3. **System extracts audio tracks** (`daily.py:211`):
```python
track_keys = [t.s3Key for t in tracks if t.type == "audio"]
```
4. **Triggers multitrack processing** (`daily.py:213-218`):
```python
process_multitrack_recording.delay(
bucket_name=bucket_name, # reflector-dailyco-local
room_name=room_name, # daily-20251020193458
recording_id=recording_id, # 7443ee0a-dab1-40eb-b316-33d6c0d5ff88
track_keys=track_keys # Only audio s3Keys
)
```
### Key Advantage: No Ambiguity
Even though multiple meetings may share the same S3 bucket/folder (`monadical/`), **there's no ambiguity** because:
- Each webhook payload contains the exact `s3Key` list for that specific `recording_id`
- No need to scan folders or guess which files belong together
- Each track's s3Key includes the room timestamp subfolder (e.g., `daily-20251020193458/`)
The room name includes timestamp (`daily-20251020193458`) to keep recordings organized, but **the webhook's explicit track list is what prevents mixing files from different meetings**.
### Track Timeline Extraction
Daily.co provides timing information in two places:
**1. PyAV WebM Metadata (current approach)**:
```python
# Read from WebM container stream metadata
stream.start_time = 8.130s # Meeting-relative timing
```
**2. Filename Timestamps (alternative approach, commit 3bae9076)**:
```
Filename format: {recording_start_ts}-{uuid}-cam-audio-{track_start_ts}.webm
Example: 1760988935484-52f7f48b-fbab-431f-9a50-87b9abfc8255-cam-audio-1760988935922.webm
Parse timestamps:
- recording_start_ts: 1760988935484 (Unix ms)
- track_start_ts: 1760988935922 (Unix ms)
- offset: (1760988935922 - 1760988935484) / 1000 = 0.438s
```
**Time Difference (PyAV vs Filename)**:
```
Track 0:
Filename offset: 438ms
PyAV metadata: 229ms
Difference: 209ms
Track 1:
Filename offset: 8339ms
PyAV metadata: 8130ms
Difference: 209ms
```
**Consistent 209ms delta** suggests network/encoding delay between file upload initiation (filename) and actual audio stream start (metadata).
**Current implementation uses PyAV metadata** because:
- More accurate (represents when audio actually started)
- Padding BEFORE transcription produces correct Whisper timestamps automatically
- No manual offset adjustment needed during transcript merge
### Why Re-encoding During Padding
Padding coincidentally involves re-encoding, which is important for Daily.co + Whisper:
**Problem:** Daily.co skips frames in recordings when microphone is muted or paused
- WebM containers have gaps where audio frames should be
- Whisper doesn't understand these gaps and produces incorrect timestamps
- Example: 5s of audio with 2s muted → file has frames only for 3s, Whisper thinks duration is 3s
**Solution:** Re-encoding via PyAV filter graph (`adelay` + `aresample`)
- Restores missing frames as silence
- Produces continuous audio stream without gaps
- Whisper now sees correct duration and produces accurate timestamps
**Why combined with padding:**
- Already re-encoding for padding (adding initial silence)
- More performant to do both operations in single PyAV pipeline
- Padded values needed for mixdown anyway (creating final MP3)
Implementation: `main_multitrack_pipeline.py:_apply_audio_padding_streaming()`
---
## Whereby (SQS-based)
Whereby uses **AWS SQS** (via S3 notifications) to notify Reflector when files are uploaded.
### How It Works
1. **Whereby uploads recording** to S3
2. **S3 sends notification** to SQS queue (one notification per file)
3. **Reflector polls SQS queue** (`worker/process.py:process_messages()`)
4. **System processes single file** (`worker/process.py:process_recording()`)
### Key Difference from Daily.co
**Whereby (SQS):** System receives S3 notification "file X was created" - only knows about one file at a time, would need to scan folder to find related files
**Daily.co (Webhook):** Daily explicitly tells system which files belong together in the webhook payload
---

View File

@@ -1,233 +0,0 @@
# Reflector Webhook Documentation
## Overview
Reflector supports webhook notifications to notify external systems when transcript processing is completed. Webhooks can be configured per room and are triggered automatically after a transcript is successfully processed.
## Configuration
Webhooks are configured at the room level with two fields:
- `webhook_url`: The HTTPS endpoint to receive webhook notifications
- `webhook_secret`: Optional secret key for HMAC signature verification (auto-generated if not provided)
## Events
### `transcript.completed`
Triggered when a transcript has been fully processed, including transcription, diarization, summarization, topic detection and calendar event integration.
### `test`
A test event that can be triggered manually to verify webhook configuration.
## Webhook Request Format
### Headers
All webhook requests include the following headers:
| Header | Description | Example |
|--------|-------------|---------|
| `Content-Type` | Always `application/json` | `application/json` |
| `User-Agent` | Identifies Reflector as the source | `Reflector-Webhook/1.0` |
| `X-Webhook-Event` | The event type | `transcript.completed` or `test` |
| `X-Webhook-Retry` | Current retry attempt number | `0`, `1`, `2`... |
| `X-Webhook-Signature` | HMAC signature (if secret configured) | `t=1735306800,v1=abc123...` |
### Signature Verification
If a webhook secret is configured, Reflector includes an HMAC-SHA256 signature in the `X-Webhook-Signature` header to verify the webhook authenticity.
The signature format is: `t={timestamp},v1={signature}`
To verify the signature:
1. Extract the timestamp and signature from the header
2. Create the signed payload: `{timestamp}.{request_body}`
3. Compute HMAC-SHA256 of the signed payload using your webhook secret
4. Compare the computed signature with the received signature
Example verification (Python):
```python
import hmac
import hashlib
def verify_webhook_signature(payload: bytes, signature_header: str, secret: str) -> bool:
# Parse header: "t=1735306800,v1=abc123..."
parts = dict(part.split("=") for part in signature_header.split(","))
timestamp = parts["t"]
received_signature = parts["v1"]
# Create signed payload
signed_payload = f"{timestamp}.{payload.decode('utf-8')}"
# Compute expected signature
expected_signature = hmac.new(
secret.encode("utf-8"),
signed_payload.encode("utf-8"),
hashlib.sha256
).hexdigest()
# Compare signatures
return hmac.compare_digest(expected_signature, received_signature)
```
## Event Payloads
### `transcript.completed` Event
This event includes a convenient URL for accessing the transcript:
- `frontend_url`: Direct link to view the transcript in the web interface
```json
{
"event": "transcript.completed",
"event_id": "transcript.completed-abc-123-def-456",
"timestamp": "2025-08-27T12:34:56.789012Z",
"transcript": {
"id": "abc-123-def-456",
"room_id": "room-789",
"created_at": "2025-08-27T12:00:00Z",
"duration": 1800.5,
"title": "Q3 Product Planning Meeting",
"short_summary": "Team discussed Q3 product roadmap, prioritizing mobile app features and API improvements.",
"long_summary": "The product team met to finalize the Q3 roadmap. Key decisions included...",
"webvtt": "WEBVTT\n\n00:00:00.000 --> 00:00:05.000\n<v Speaker 1>Welcome everyone to today's meeting...",
"topics": [
{
"title": "Introduction and Agenda",
"summary": "Meeting kickoff with agenda review",
"timestamp": 0.0,
"duration": 120.0,
"webvtt": "WEBVTT\n\n00:00:00.000 --> 00:00:05.000\n<v Speaker 1>Welcome everyone..."
},
{
"title": "Mobile App Features Discussion",
"summary": "Team reviewed proposed mobile app features for Q3",
"timestamp": 120.0,
"duration": 600.0,
"webvtt": "WEBVTT\n\n00:02:00.000 --> 00:02:10.000\n<v Speaker 2>Let's talk about the mobile app..."
}
],
"participants": [
{
"id": "participant-1",
"name": "John Doe",
"speaker": "Speaker 1"
},
{
"id": "participant-2",
"name": "Jane Smith",
"speaker": "Speaker 2"
}
],
"source_language": "en",
"target_language": "en",
"status": "completed",
"frontend_url": "https://app.reflector.com/transcripts/abc-123-def-456"
},
"room": {
"id": "room-789",
"name": "Product Team Room"
},
"calendar_event": {
"id": "calendar-event-123",
"ics_uid": "event-123",
"title": "Q3 Product Planning Meeting",
"start_time": "2025-08-27T12:00:00Z",
"end_time": "2025-08-27T12:30:00Z",
"description": "Team discussed Q3 product roadmap, prioritizing mobile app features and API improvements.",
"location": "Conference Room 1",
"attendees": [
{
"id": "participant-1",
"name": "John Doe",
"speaker": "Speaker 1"
},
{
"id": "participant-2",
"name": "Jane Smith",
"speaker": "Speaker 2"
}
]
}
}
```
### `test` Event
```json
{
"event": "test",
"event_id": "test.2025-08-27T12:34:56.789012Z",
"timestamp": "2025-08-27T12:34:56.789012Z",
"message": "This is a test webhook from Reflector",
"room": {
"id": "room-789",
"name": "Product Team Room"
}
}
```
## Retry Policy
Webhooks are delivered with automatic retry logic to handle transient failures. When a webhook delivery fails due to server errors or network issues, Reflector will automatically retry the delivery multiple times over an extended period.
### Retry Mechanism
Reflector implements an exponential backoff strategy for webhook retries:
- **Initial retry delay**: 60 seconds after the first failure
- **Exponential backoff**: Each subsequent retry waits approximately twice as long as the previous one
- **Maximum retry interval**: 1 hour (backoff is capped at this duration)
- **Maximum retry attempts**: 30 attempts total
- **Total retry duration**: Retries continue for approximately 24 hours
### How Retries Work
When a webhook fails, Reflector will:
1. Wait 60 seconds, then retry (attempt #1)
2. If it fails again, wait ~2 minutes, then retry (attempt #2)
3. Continue doubling the wait time up to a maximum of 1 hour between attempts
4. Keep retrying at 1-hour intervals until successful or 30 attempts are exhausted
The `X-Webhook-Retry` header indicates the current retry attempt number (0 for the initial attempt, 1 for first retry, etc.), allowing your endpoint to track retry attempts.
### Retry Behavior by HTTP Status Code
| Status Code | Behavior |
|-------------|----------|
| 2xx (Success) | No retry, webhook marked as delivered |
| 4xx (Client Error) | No retry, request is considered permanently failed |
| 5xx (Server Error) | Automatic retry with exponential backoff |
| Network/Timeout Error | Automatic retry with exponential backoff |
**Important Notes:**
- Webhooks timeout after 30 seconds. If your endpoint takes longer to respond, it will be considered a timeout error and retried.
- During the retry period (~24 hours), you may receive the same webhook multiple times if your endpoint experiences intermittent failures.
- There is no mechanism to manually retry failed webhooks after the retry period expires.
## Testing Webhooks
You can test your webhook configuration before processing transcripts:
```http
POST /v1/rooms/{room_id}/webhook/test
```
Response:
```json
{
"success": true,
"status_code": 200,
"message": "Webhook test successful",
"response_preview": "OK"
}
```
Or in case of failure:
```json
{
"success": false,
"error": "Webhook request timed out (10 seconds)"
}
```

View File

@@ -24,20 +24,19 @@ AUTH_JWT_AUDIENCE=
## Using serverless modal.com (require reflector-gpu-modal deployed)
#TRANSCRIPT_BACKEND=modal
#TRANSCRIPT_URL=https://xxxxx--reflector-transcriber-web.modal.run
#TRANSLATE_URL=https://xxxxx--reflector-translator-web.modal.run
#TRANSCRIPT_MODAL_API_KEY=xxxxx
TRANSCRIPT_BACKEND=modal
TRANSCRIPT_URL=https://monadical-sas--reflector-transcriber-parakeet-web.modal.run
TRANSCRIPT_URL=https://monadical-sas--reflector-transcriber-web.modal.run
TRANSCRIPT_MODAL_API_KEY=
## =======================================================
## Translation backend
## Transcription backend
##
## Only available in modal atm
## =======================================================
TRANSLATION_BACKEND=modal
TRANSLATE_URL=https://monadical-sas--reflector-translator-web.modal.run
#TRANSLATION_MODAL_API_KEY=xxxxx
## =======================================================
## LLM backend
@@ -60,9 +59,7 @@ LLM_API_KEY=sk-
## To allow diarization, you need to expose expose the files to be dowloded by the pipeline
## =======================================================
DIARIZATION_ENABLED=false
DIARIZATION_BACKEND=modal
DIARIZATION_URL=https://monadical-sas--reflector-diarizer-web.modal.run
#DIARIZATION_MODAL_API_KEY=xxxxx
## =======================================================
@@ -71,30 +68,3 @@ DIARIZATION_URL=https://monadical-sas--reflector-diarizer-web.modal.run
## Sentry DSN configuration
#SENTRY_DSN=
## =======================================================
## Video Platform Configuration
## =======================================================
## Whereby
#WHEREBY_API_KEY=your-whereby-api-key
#WHEREBY_WEBHOOK_SECRET=your-whereby-webhook-secret
#WHEREBY_STORAGE_AWS_ACCESS_KEY_ID=your-aws-key
#WHEREBY_STORAGE_AWS_SECRET_ACCESS_KEY=your-aws-secret
#AWS_PROCESS_RECORDING_QUEUE_URL=https://sqs.us-west-2.amazonaws.com/...
## Daily.co
#DAILY_API_KEY=your-daily-api-key
#DAILY_WEBHOOK_SECRET=your-daily-webhook-secret
#DAILY_SUBDOMAIN=your-subdomain
#DAILY_WEBHOOK_UUID= # Auto-populated by recreate_daily_webhook.py script
#DAILYCO_STORAGE_AWS_ROLE_ARN=... # IAM role ARN for Daily.co S3 access
#DAILYCO_STORAGE_AWS_BUCKET_NAME=reflector-dailyco
#DAILYCO_STORAGE_AWS_REGION=us-west-2
## Whereby (optional separate bucket)
#WHEREBY_STORAGE_AWS_BUCKET_NAME=reflector-whereby
#WHEREBY_STORAGE_AWS_REGION=us-east-1
## Platform Configuration
#DEFAULT_VIDEO_PLATFORM=whereby # Default platform for new rooms

View File

@@ -0,0 +1,82 @@
# Reflector GPU implementation - Transcription and LLM
This repository hold an API for the GPU implementation of the Reflector API service,
and use [Modal.com](https://modal.com)
- `reflector_diarizer.py` - Diarization API
- `reflector_transcriber.py` - Transcription API
- `reflector_translator.py` - Translation API
## Modal.com deployment
Create a modal secret, and name it `reflector-gpu`.
It should contain an `REFLECTOR_APIKEY` environment variable with a value.
The deployment is done using [Modal.com](https://modal.com) service.
```
$ modal deploy reflector_transcriber.py
...
└── 🔨 Created web => https://xxxx--reflector-transcriber-web.modal.run
$ modal deploy reflector_llm.py
...
└── 🔨 Created web => https://xxxx--reflector-llm-web.modal.run
```
Then in your reflector api configuration `.env`, you can set theses keys:
```
TRANSCRIPT_BACKEND=modal
TRANSCRIPT_URL=https://xxxx--reflector-transcriber-web.modal.run
TRANSCRIPT_MODAL_API_KEY=REFLECTOR_APIKEY
LLM_BACKEND=modal
LLM_URL=https://xxxx--reflector-llm-web.modal.run
LLM_MODAL_API_KEY=REFLECTOR_APIKEY
```
## API
Authentication must be passed with the `Authorization` header, using the `bearer` scheme.
```
Authorization: bearer <REFLECTOR_APIKEY>
```
### LLM
`POST /llm`
**request**
```
{
"prompt": "xxx"
}
```
**response**
```
{
"text": "xxx completed"
}
```
### Transcription
`POST /transcribe`
**request** (multipart/form-data)
- `file` - audio file
- `language` - language code (e.g. `en`)
**response**
```
{
"text": "xxx",
"words": [
{"text": "xxx", "start": 0.0, "end": 1.0}
]
}
```

View File

@@ -0,0 +1,187 @@
"""
Reflector GPU backend - diarizer
===================================
"""
import os
import modal.gpu
from modal import App, Image, Secret, asgi_app, enter, method
from pydantic import BaseModel
PYANNOTE_MODEL_NAME: str = "pyannote/speaker-diarization-3.1"
MODEL_DIR = "/root/diarization_models"
app = App(name="reflector-diarizer")
def migrate_cache_llm():
"""
XXX The cache for model files in Transformers v4.22.0 has been updated.
Migrating your old cache. This is a one-time only operation. You can
interrupt this and resume the migration later on by calling
`transformers.utils.move_cache()`.
"""
from transformers.utils.hub import move_cache
print("Moving LLM cache")
move_cache(cache_dir=MODEL_DIR, new_cache_dir=MODEL_DIR)
print("LLM cache moved")
def download_pyannote_audio():
from pyannote.audio import Pipeline
Pipeline.from_pretrained(
PYANNOTE_MODEL_NAME,
cache_dir=MODEL_DIR,
use_auth_token=os.environ["HF_TOKEN"],
)
diarizer_image = (
Image.debian_slim(python_version="3.10.8")
.pip_install(
"pyannote.audio==3.1.0",
"requests",
"onnx",
"torchaudio",
"onnxruntime-gpu",
"torch==2.0.0",
"transformers==4.34.0",
"sentencepiece",
"protobuf",
"numpy",
"huggingface_hub",
"hf-transfer",
)
.run_function(
download_pyannote_audio, secrets=[Secret.from_name("my-huggingface-secret")]
)
.run_function(migrate_cache_llm)
.env(
{
"LD_LIBRARY_PATH": (
"/usr/local/lib/python3.10/site-packages/nvidia/cudnn/lib/:"
"/opt/conda/lib/python3.10/site-packages/nvidia/cublas/lib/"
)
}
)
)
@app.cls(
gpu=modal.gpu.A100(size="40GB"),
timeout=60 * 30,
scaledown_window=60,
allow_concurrent_inputs=1,
image=diarizer_image,
)
class Diarizer:
@enter()
def enter(self):
import torch
from pyannote.audio import Pipeline
self.use_gpu = torch.cuda.is_available()
self.device = "cuda" if self.use_gpu else "cpu"
self.diarization_pipeline = Pipeline.from_pretrained(
PYANNOTE_MODEL_NAME, cache_dir=MODEL_DIR
)
self.diarization_pipeline.to(torch.device(self.device))
@method()
def diarize(self, audio_data: str, audio_suffix: str, timestamp: float):
import tempfile
import torchaudio
with tempfile.NamedTemporaryFile("wb+", suffix=f".{audio_suffix}") as fp:
fp.write(audio_data)
print("Diarizing audio")
waveform, sample_rate = torchaudio.load(fp.name)
diarization = self.diarization_pipeline(
{"waveform": waveform, "sample_rate": sample_rate}
)
words = []
for diarization_segment, _, speaker in diarization.itertracks(
yield_label=True
):
words.append(
{
"start": round(timestamp + diarization_segment.start, 3),
"end": round(timestamp + diarization_segment.end, 3),
"speaker": int(speaker[-2:]),
}
)
print("Diarization complete")
return {"diarization": words}
# -------------------------------------------------------------------
# Web API
# -------------------------------------------------------------------
@app.function(
timeout=60 * 10,
scaledown_window=60 * 3,
allow_concurrent_inputs=40,
secrets=[
Secret.from_name("reflector-gpu"),
],
image=diarizer_image,
)
@asgi_app()
def web():
import requests
from fastapi import Depends, FastAPI, HTTPException, status
from fastapi.security import OAuth2PasswordBearer
diarizerstub = Diarizer()
app = FastAPI()
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
def apikey_auth(apikey: str = Depends(oauth2_scheme)):
if apikey != os.environ["REFLECTOR_GPU_APIKEY"]:
raise HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail="Invalid API key",
headers={"WWW-Authenticate": "Bearer"},
)
def validate_audio_file(audio_file_url: str):
# Check if the audio file exists
response = requests.head(audio_file_url, allow_redirects=True)
if response.status_code == 404:
raise HTTPException(
status_code=response.status_code,
detail="The audio file does not exist.",
)
class DiarizationResponse(BaseModel):
result: dict
@app.post(
"/diarize", dependencies=[Depends(apikey_auth), Depends(validate_audio_file)]
)
def diarize(
audio_file_url: str, timestamp: float = 0.0
) -> HTTPException | DiarizationResponse:
# Currently the uploaded files are in mp3 format
audio_suffix = "mp3"
print("Downloading audio file")
response = requests.get(audio_file_url, allow_redirects=True)
print("Audio file downloaded successfully")
func = diarizerstub.diarize.spawn(
audio_data=response.content, audio_suffix=audio_suffix, timestamp=timestamp
)
result = func.get()
return result
return app

View File

@@ -0,0 +1,161 @@
import os
import tempfile
import threading
import modal
from pydantic import BaseModel
MODELS_DIR = "/models"
MODEL_NAME = "large-v2"
MODEL_COMPUTE_TYPE: str = "float16"
MODEL_NUM_WORKERS: int = 1
MINUTES = 60 # seconds
volume = modal.Volume.from_name("models", create_if_missing=True)
app = modal.App("reflector-transcriber")
def download_model():
from faster_whisper import download_model
volume.reload()
download_model(MODEL_NAME, cache_dir=MODELS_DIR)
volume.commit()
image = (
modal.Image.debian_slim(python_version="3.12")
.pip_install(
"huggingface_hub==0.27.1",
"hf-transfer==0.1.9",
"torch==2.5.1",
"faster-whisper==1.1.1",
)
.env(
{
"HF_HUB_ENABLE_HF_TRANSFER": "1",
"LD_LIBRARY_PATH": (
"/usr/local/lib/python3.12/site-packages/nvidia/cudnn/lib/:"
"/opt/conda/lib/python3.12/site-packages/nvidia/cublas/lib/"
),
}
)
.run_function(download_model, volumes={MODELS_DIR: volume})
)
@app.cls(
gpu="A10G",
timeout=5 * MINUTES,
scaledown_window=5 * MINUTES,
allow_concurrent_inputs=6,
image=image,
volumes={MODELS_DIR: volume},
)
class Transcriber:
@modal.enter()
def enter(self):
import faster_whisper
import torch
self.lock = threading.Lock()
self.use_gpu = torch.cuda.is_available()
self.device = "cuda" if self.use_gpu else "cpu"
self.model = faster_whisper.WhisperModel(
MODEL_NAME,
device=self.device,
compute_type=MODEL_COMPUTE_TYPE,
num_workers=MODEL_NUM_WORKERS,
download_root=MODELS_DIR,
local_files_only=True,
)
@modal.method()
def transcribe_segment(
self,
audio_data: str,
audio_suffix: str,
language: str,
):
with tempfile.NamedTemporaryFile("wb+", suffix=f".{audio_suffix}") as fp:
fp.write(audio_data)
with self.lock:
segments, _ = self.model.transcribe(
fp.name,
language=language,
beam_size=5,
word_timestamps=True,
vad_filter=True,
vad_parameters={"min_silence_duration_ms": 500},
)
segments = list(segments)
text = "".join(segment.text for segment in segments)
words = [
{"word": word.word, "start": word.start, "end": word.end}
for segment in segments
for word in segment.words
]
return {"text": text, "words": words}
@app.function(
scaledown_window=60,
timeout=60,
allow_concurrent_inputs=40,
secrets=[
modal.Secret.from_name("reflector-gpu"),
],
volumes={MODELS_DIR: volume},
)
@modal.asgi_app()
def web():
from fastapi import Body, Depends, FastAPI, HTTPException, UploadFile, status
from fastapi.security import OAuth2PasswordBearer
from typing_extensions import Annotated
transcriber = Transcriber()
app = FastAPI()
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token")
supported_file_types = ["mp3", "mp4", "mpeg", "mpga", "m4a", "wav", "webm"]
def apikey_auth(apikey: str = Depends(oauth2_scheme)):
if apikey != os.environ["REFLECTOR_GPU_APIKEY"]:
raise HTTPException(
status_code=status.HTTP_401_UNAUTHORIZED,
detail="Invalid API key",
headers={"WWW-Authenticate": "Bearer"},
)
class TranscriptResponse(BaseModel):
result: dict
@app.post("/v1/audio/transcriptions", dependencies=[Depends(apikey_auth)])
def transcribe(
file: UploadFile,
model: str = "whisper-1",
language: Annotated[str, Body(...)] = "en",
) -> TranscriptResponse:
audio_data = file.file.read()
audio_suffix = file.filename.split(".")[-1]
assert audio_suffix in supported_file_types
func = transcriber.transcribe_segment.spawn(
audio_data=audio_data,
audio_suffix=audio_suffix,
language=language,
)
result = func.get()
return result
return app

View File

@@ -1,3 +1 @@
Generic single-database configuration.
Both data migrations and schema migrations must be in migrations.
Generic single-database configuration.

View File

@@ -1,36 +0,0 @@
"""Add webhook fields to rooms
Revision ID: 0194f65cd6d3
Revises: 5a8907fd1d78
Create Date: 2025-08-27 09:03:19.610995
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "0194f65cd6d3"
down_revision: Union[str, None] = "5a8907fd1d78"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.add_column(sa.Column("webhook_url", sa.String(), nullable=True))
batch_op.add_column(sa.Column("webhook_secret", sa.String(), nullable=True))
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.drop_column("webhook_secret")
batch_op.drop_column("webhook_url")
# ### end Alembic commands ###

View File

@@ -1,26 +0,0 @@
"""add_action_items
Revision ID: 05f8688d6895
Revises: bbafedfa510c
Create Date: 2025-12-12 11:57:50.209658
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "05f8688d6895"
down_revision: Union[str, None] = "bbafedfa510c"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
op.add_column("transcript", sa.Column("action_items", sa.JSON(), nullable=True))
def downgrade() -> None:
op.drop_column("transcript", "action_items")

View File

@@ -1,64 +0,0 @@
"""add_long_summary_to_search_vector
Revision ID: 0ab2d7ffaa16
Revises: b1c33bd09963
Create Date: 2025-08-15 13:27:52.680211
"""
from typing import Sequence, Union
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "0ab2d7ffaa16"
down_revision: Union[str, None] = "b1c33bd09963"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# Drop the existing search vector column and index
op.drop_index("idx_transcript_search_vector_en", table_name="transcript")
op.drop_column("transcript", "search_vector_en")
# Recreate the search vector column with long_summary included
op.execute("""
ALTER TABLE transcript ADD COLUMN search_vector_en tsvector
GENERATED ALWAYS AS (
setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
setweight(to_tsvector('english', coalesce(long_summary, '')), 'B') ||
setweight(to_tsvector('english', coalesce(webvtt, '')), 'C')
) STORED
""")
# Recreate the GIN index for the search vector
op.create_index(
"idx_transcript_search_vector_en",
"transcript",
["search_vector_en"],
postgresql_using="gin",
)
def downgrade() -> None:
# Drop the updated search vector column and index
op.drop_index("idx_transcript_search_vector_en", table_name="transcript")
op.drop_column("transcript", "search_vector_en")
# Recreate the original search vector column without long_summary
op.execute("""
ALTER TABLE transcript ADD COLUMN search_vector_en tsvector
GENERATED ALWAYS AS (
setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
setweight(to_tsvector('english', coalesce(webvtt, '')), 'B')
) STORED
""")
# Recreate the GIN index for the search vector
op.create_index(
"idx_transcript_search_vector_en",
"transcript",
["search_vector_en"],
postgresql_using="gin",
)

View File

@@ -1,25 +0,0 @@
"""add_webvtt_field_to_transcript
Revision ID: 0bc0f3ff0111
Revises: b7df9609542c
Create Date: 2025-08-05 19:36:41.740957
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
revision: str = "0bc0f3ff0111"
down_revision: Union[str, None] = "b7df9609542c"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
op.add_column("transcript", sa.Column("webvtt", sa.Text(), nullable=True))
def downgrade() -> None:
op.drop_column("transcript", "webvtt")

View File

@@ -1,36 +0,0 @@
"""remove user_id from meeting table
Revision ID: 0ce521cda2ee
Revises: 6dec9fb5b46c
Create Date: 2025-09-10 12:40:55.688899
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "0ce521cda2ee"
down_revision: Union[str, None] = "6dec9fb5b46c"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.drop_column("user_id")
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.add_column(
sa.Column("user_id", sa.VARCHAR(), autoincrement=False, nullable=True)
)
# ### end Alembic commands ###

View File

@@ -1,46 +0,0 @@
"""add_full_text_search
Revision ID: 116b2f287eab
Revises: 0bc0f3ff0111
Create Date: 2025-08-07 11:27:38.473517
"""
from typing import Sequence, Union
from alembic import op
revision: str = "116b2f287eab"
down_revision: Union[str, None] = "0bc0f3ff0111"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
conn = op.get_bind()
if conn.dialect.name != "postgresql":
return
op.execute("""
ALTER TABLE transcript ADD COLUMN search_vector_en tsvector
GENERATED ALWAYS AS (
setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
setweight(to_tsvector('english', coalesce(webvtt, '')), 'B')
) STORED
""")
op.create_index(
"idx_transcript_search_vector_en",
"transcript",
["search_vector_en"],
postgresql_using="gin",
)
def downgrade() -> None:
conn = op.get_bind()
if conn.dialect.name != "postgresql":
return
op.drop_index("idx_transcript_search_vector_en", table_name="transcript")
op.drop_column("transcript", "search_vector_en")

View File

@@ -1,50 +0,0 @@
"""add_platform_support
Revision ID: 1e49625677e4
Revises: 9e3f7b2a4c8e
Create Date: 2025-10-08 13:17:29.943612
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "1e49625677e4"
down_revision: Union[str, None] = "9e3f7b2a4c8e"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
"""Add platform field with default 'whereby' for backward compatibility."""
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.add_column(
sa.Column(
"platform",
sa.String(),
nullable=True,
server_default=None,
)
)
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.add_column(
sa.Column(
"platform",
sa.String(),
nullable=False,
server_default="whereby",
)
)
def downgrade() -> None:
"""Remove platform field."""
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.drop_column("platform")
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.drop_column("platform")

View File

@@ -1,32 +0,0 @@
"""clean up orphaned room_id references in meeting table
Revision ID: 2ae3db106d4e
Revises: def1b5867d4c
Create Date: 2025-09-11 10:35:15.759967
"""
from typing import Sequence, Union
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "2ae3db106d4e"
down_revision: Union[str, None] = "def1b5867d4c"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# Set room_id to NULL for meetings that reference non-existent rooms
op.execute("""
UPDATE meeting
SET room_id = NULL
WHERE room_id IS NOT NULL
AND room_id NOT IN (SELECT id FROM room WHERE id IS NOT NULL)
""")
def downgrade() -> None:
# Cannot restore orphaned references - no operation needed
pass

View File

@@ -1,79 +0,0 @@
"""add daily participant session table with immutable left_at
Revision ID: 2b92a1b03caa
Revises: f8294b31f022
Create Date: 2025-11-13 20:29:30.486577
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "2b92a1b03caa"
down_revision: Union[str, None] = "f8294b31f022"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# Create table
op.create_table(
"daily_participant_session",
sa.Column("id", sa.String(), nullable=False),
sa.Column("meeting_id", sa.String(), nullable=False),
sa.Column("room_id", sa.String(), nullable=False),
sa.Column("session_id", sa.String(), nullable=False),
sa.Column("user_id", sa.String(), nullable=True),
sa.Column("user_name", sa.String(), nullable=False),
sa.Column("joined_at", sa.DateTime(timezone=True), nullable=False),
sa.Column("left_at", sa.DateTime(timezone=True), nullable=True),
sa.ForeignKeyConstraint(["meeting_id"], ["meeting.id"], ondelete="CASCADE"),
sa.ForeignKeyConstraint(["room_id"], ["room.id"], ondelete="CASCADE"),
sa.PrimaryKeyConstraint("id"),
)
with op.batch_alter_table("daily_participant_session", schema=None) as batch_op:
batch_op.create_index(
"idx_daily_session_meeting_left", ["meeting_id", "left_at"], unique=False
)
batch_op.create_index("idx_daily_session_room", ["room_id"], unique=False)
# Create trigger function to prevent left_at from being updated once set
op.execute("""
CREATE OR REPLACE FUNCTION prevent_left_at_update()
RETURNS TRIGGER AS $$
BEGIN
IF OLD.left_at IS NOT NULL THEN
RAISE EXCEPTION 'left_at is immutable once set';
END IF;
RETURN NEW;
END;
$$ LANGUAGE plpgsql;
""")
# Create trigger
op.execute("""
CREATE TRIGGER prevent_left_at_update_trigger
BEFORE UPDATE ON daily_participant_session
FOR EACH ROW
EXECUTE FUNCTION prevent_left_at_update();
""")
def downgrade() -> None:
# Drop trigger
op.execute(
"DROP TRIGGER IF EXISTS prevent_left_at_update_trigger ON daily_participant_session;"
)
# Drop trigger function
op.execute("DROP FUNCTION IF EXISTS prevent_left_at_update();")
# Drop indexes and table
with op.batch_alter_table("daily_participant_session", schema=None) as batch_op:
batch_op.drop_index("idx_daily_session_room")
batch_op.drop_index("idx_daily_session_meeting_left")
op.drop_table("daily_participant_session")

View File

@@ -1,50 +0,0 @@
"""add cascade delete to meeting consent foreign key
Revision ID: 5a8907fd1d78
Revises: 0ab2d7ffaa16
Create Date: 2025-08-26 17:26:50.945491
"""
from typing import Sequence, Union
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "5a8907fd1d78"
down_revision: Union[str, None] = "0ab2d7ffaa16"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting_consent", schema=None) as batch_op:
batch_op.drop_constraint(
batch_op.f("meeting_consent_meeting_id_fkey"), type_="foreignkey"
)
batch_op.create_foreign_key(
batch_op.f("meeting_consent_meeting_id_fkey"),
"meeting",
["meeting_id"],
["id"],
ondelete="CASCADE",
)
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting_consent", schema=None) as batch_op:
batch_op.drop_constraint(
batch_op.f("meeting_consent_meeting_id_fkey"), type_="foreignkey"
)
batch_op.create_foreign_key(
batch_op.f("meeting_consent_meeting_id_fkey"),
"meeting",
["meeting_id"],
["id"],
)
# ### end Alembic commands ###

View File

@@ -1,30 +0,0 @@
"""Make room platform non-nullable with dynamic default
Revision ID: 5d6b9df9b045
Revises: 2b92a1b03caa
Create Date: 2025-11-21 13:22:25.756584
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "5d6b9df9b045"
down_revision: Union[str, None] = "2b92a1b03caa"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
op.execute("UPDATE room SET platform = 'whereby' WHERE platform IS NULL")
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.alter_column("platform", existing_type=sa.String(), nullable=False)
def downgrade() -> None:
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.alter_column("platform", existing_type=sa.String(), nullable=True)

View File

@@ -1,53 +0,0 @@
"""remove_one_active_meeting_per_room_constraint
Revision ID: 6025e9b2bef2
Revises: 2ae3db106d4e
Create Date: 2025-08-18 18:45:44.418392
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "6025e9b2bef2"
down_revision: Union[str, None] = "2ae3db106d4e"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# Remove the unique constraint that prevents multiple active meetings per room
# This is needed to support calendar integration with overlapping meetings
# Check if index exists before trying to drop it
from alembic import context
if context.get_context().dialect.name == "postgresql":
conn = op.get_bind()
result = conn.execute(
sa.text(
"SELECT 1 FROM pg_indexes WHERE indexname = 'idx_one_active_meeting_per_room'"
)
)
if result.fetchone():
op.drop_index("idx_one_active_meeting_per_room", table_name="meeting")
else:
# For SQLite, just try to drop it
try:
op.drop_index("idx_one_active_meeting_per_room", table_name="meeting")
except:
pass
def downgrade() -> None:
# Restore the unique constraint
op.create_index(
"idx_one_active_meeting_per_room",
"meeting",
["room_id"],
unique=True,
postgresql_where=sa.text("is_active = true"),
sqlite_where=sa.text("is_active = 1"),
)

View File

@@ -1,28 +0,0 @@
"""webhook url and secret null by default
Revision ID: 61882a919591
Revises: 0194f65cd6d3
Create Date: 2025-08-29 11:46:36.738091
"""
from typing import Sequence, Union
# revision identifiers, used by Alembic.
revision: str = "61882a919591"
down_revision: Union[str, None] = "0194f65cd6d3"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
pass
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
pass
# ### end Alembic commands ###

View File

@@ -32,7 +32,7 @@ def upgrade() -> None:
sa.Column("user_id", sa.String(), nullable=True),
sa.Column("room_id", sa.String(), nullable=True),
sa.Column(
"is_locked", sa.Boolean(), server_default=sa.text("false"), nullable=False
"is_locked", sa.Boolean(), server_default=sa.text("0"), nullable=False
),
sa.Column("room_mode", sa.String(), server_default="normal", nullable=False),
sa.Column(
@@ -53,15 +53,12 @@ def upgrade() -> None:
sa.Column("user_id", sa.String(), nullable=False),
sa.Column("created_at", sa.DateTime(), nullable=False),
sa.Column(
"zulip_auto_post",
sa.Boolean(),
server_default=sa.text("false"),
nullable=False,
"zulip_auto_post", sa.Boolean(), server_default=sa.text("0"), nullable=False
),
sa.Column("zulip_stream", sa.String(), nullable=True),
sa.Column("zulip_topic", sa.String(), nullable=True),
sa.Column(
"is_locked", sa.Boolean(), server_default=sa.text("false"), nullable=False
"is_locked", sa.Boolean(), server_default=sa.text("0"), nullable=False
),
sa.Column("room_mode", sa.String(), server_default="normal", nullable=False),
sa.Column(

View File

@@ -1,35 +0,0 @@
"""make meeting room_id required and add foreign key
Revision ID: 6dec9fb5b46c
Revises: 61882a919591
Create Date: 2025-09-10 10:47:06.006819
"""
from typing import Sequence, Union
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "6dec9fb5b46c"
down_revision: Union[str, None] = "61882a919591"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.create_foreign_key(
None, "room", ["room_id"], ["id"], ondelete="CASCADE"
)
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.drop_constraint("meeting_room_id_fkey", type_="foreignkey")
# ### end Alembic commands ###

View File

@@ -20,14 +20,11 @@ depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
sourcekind_enum = sa.Enum("room", "live", "file", name="sourcekind")
sourcekind_enum.create(op.get_bind())
op.add_column(
"transcript",
sa.Column(
"source_kind",
sourcekind_enum,
sa.Enum("ROOM", "LIVE", "FILE", name="sourcekind"),
nullable=True,
),
)
@@ -46,8 +43,6 @@ def upgrade() -> None:
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.drop_column("transcript", "source_kind")
sourcekind_enum = sa.Enum(name="sourcekind")
sourcekind_enum.drop(op.get_bind())
# ### end Alembic commands ###

View File

@@ -0,0 +1,54 @@
"""dailyco platform
Revision ID: 7e47155afd51
Revises: b7df9609542c
Create Date: 2025-08-04 11:14:19.663115
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "7e47155afd51"
down_revision: Union[str, None] = "b7df9609542c"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.add_column(
sa.Column("platform", sa.String(), server_default="whereby", nullable=False)
)
batch_op.drop_index(
batch_op.f("idx_one_active_meeting_per_room"),
sqlite_where=sa.text("is_active = 1"),
)
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.add_column(
sa.Column("platform", sa.String(), server_default="whereby", nullable=False)
)
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.drop_column("platform")
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.create_index(
batch_op.f("idx_one_active_meeting_per_room"),
["room_id"],
unique=1,
sqlite_where=sa.text("is_active = 1"),
)
batch_op.drop_column("platform")
# ### end Alembic commands ###

View File

@@ -1,106 +0,0 @@
"""populate_webvtt_from_topics
Revision ID: 8120ebc75366
Revises: 116b2f287eab
Create Date: 2025-08-11 19:11:01.316947
"""
import json
from typing import Sequence, Union
from alembic import op
from sqlalchemy import text
# revision identifiers, used by Alembic.
revision: str = "8120ebc75366"
down_revision: Union[str, None] = "116b2f287eab"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def topics_to_webvtt(topics):
"""Convert topics list to WebVTT format string."""
if not topics:
return None
lines = ["WEBVTT", ""]
for topic in topics:
start_time = format_timestamp(topic.get("start"))
end_time = format_timestamp(topic.get("end"))
text = topic.get("text", "").strip()
if start_time and end_time and text:
lines.append(f"{start_time} --> {end_time}")
lines.append(text)
lines.append("")
return "\n".join(lines).strip()
def format_timestamp(seconds):
"""Format seconds to WebVTT timestamp format (HH:MM:SS.mmm)."""
if seconds is None:
return None
hours = int(seconds // 3600)
minutes = int((seconds % 3600) // 60)
secs = seconds % 60
return f"{hours:02d}:{minutes:02d}:{secs:06.3f}"
def upgrade() -> None:
"""Populate WebVTT field for all transcripts with topics."""
# Get connection
connection = op.get_bind()
# Query all transcripts with topics
result = connection.execute(
text("SELECT id, topics FROM transcript WHERE topics IS NOT NULL")
)
rows = result.fetchall()
print(f"Found {len(rows)} transcripts with topics")
updated_count = 0
error_count = 0
for row in rows:
transcript_id = row[0]
topics_data = row[1]
if not topics_data:
continue
try:
# Parse JSON if it's a string
if isinstance(topics_data, str):
topics_data = json.loads(topics_data)
# Convert topics to WebVTT format
webvtt_content = topics_to_webvtt(topics_data)
if webvtt_content:
# Update the webvtt field
connection.execute(
text("UPDATE transcript SET webvtt = :webvtt WHERE id = :id"),
{"webvtt": webvtt_content, "id": transcript_id},
)
updated_count += 1
print(f"✓ Updated transcript {transcript_id}")
except Exception as e:
error_count += 1
print(f"✗ Error updating transcript {transcript_id}: {e}")
print(f"\nMigration complete!")
print(f" Updated: {updated_count}")
print(f" Errors: {error_count}")
def downgrade() -> None:
"""Clear WebVTT field for all transcripts."""
op.execute(text("UPDATE transcript SET webvtt = NULL"))

View File

@@ -22,7 +22,7 @@ def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.execute(
"UPDATE transcript SET events = "
'REPLACE(events::text, \'"event": "SUMMARY"\', \'"event": "LONG_SUMMARY"\')::json;'
'REPLACE(events, \'"event": "SUMMARY"\', \'"event": "LONG_SUMMARY"\');'
)
op.alter_column("transcript", "summary", new_column_name="long_summary")
op.add_column("transcript", sa.Column("title", sa.String(), nullable=True))
@@ -34,7 +34,7 @@ def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.execute(
"UPDATE transcript SET events = "
'REPLACE(events::text, \'"event": "LONG_SUMMARY"\', \'"event": "SUMMARY"\')::json;'
'REPLACE(events, \'"event": "LONG_SUMMARY"\', \'"event": "SUMMARY"\');'
)
with op.batch_alter_table("transcript", schema=None) as batch_op:
batch_op.alter_column("long_summary", nullable=True, new_column_name="summary")

View File

@@ -1,38 +0,0 @@
"""add user api keys
Revision ID: 9e3f7b2a4c8e
Revises: dc035ff72fd5
Create Date: 2025-10-17 00:00:00.000000
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "9e3f7b2a4c8e"
down_revision: Union[str, None] = "dc035ff72fd5"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
op.create_table(
"user_api_key",
sa.Column("id", sa.String(), nullable=False),
sa.Column("user_id", sa.String(), nullable=False),
sa.Column("key_hash", sa.String(), nullable=False),
sa.Column("name", sa.String(), nullable=True),
sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
sa.PrimaryKeyConstraint("id"),
)
with op.batch_alter_table("user_api_key", schema=None) as batch_op:
batch_op.create_index("idx_user_api_key_hash", ["key_hash"], unique=True)
batch_op.create_index("idx_user_api_key_user_id", ["user_id"], unique=False)
def downgrade() -> None:
op.drop_table("user_api_key")

View File

@@ -1,121 +0,0 @@
"""datetime timezone
Revision ID: 9f5c78d352d6
Revises: 8120ebc75366
Create Date: 2025-08-13 19:18:27.113593
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
from sqlalchemy.dialects import postgresql
# revision identifiers, used by Alembic.
revision: str = "9f5c78d352d6"
down_revision: Union[str, None] = "8120ebc75366"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.alter_column(
"start_date",
existing_type=postgresql.TIMESTAMP(),
type_=sa.DateTime(timezone=True),
existing_nullable=True,
)
batch_op.alter_column(
"end_date",
existing_type=postgresql.TIMESTAMP(),
type_=sa.DateTime(timezone=True),
existing_nullable=True,
)
with op.batch_alter_table("meeting_consent", schema=None) as batch_op:
batch_op.alter_column(
"consent_timestamp",
existing_type=postgresql.TIMESTAMP(),
type_=sa.DateTime(timezone=True),
existing_nullable=False,
)
with op.batch_alter_table("recording", schema=None) as batch_op:
batch_op.alter_column(
"recorded_at",
existing_type=postgresql.TIMESTAMP(),
type_=sa.DateTime(timezone=True),
existing_nullable=False,
)
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.alter_column(
"created_at",
existing_type=postgresql.TIMESTAMP(),
type_=sa.DateTime(timezone=True),
existing_nullable=False,
)
with op.batch_alter_table("transcript", schema=None) as batch_op:
batch_op.alter_column(
"created_at",
existing_type=postgresql.TIMESTAMP(),
type_=sa.DateTime(timezone=True),
existing_nullable=True,
)
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("transcript", schema=None) as batch_op:
batch_op.alter_column(
"created_at",
existing_type=sa.DateTime(timezone=True),
type_=postgresql.TIMESTAMP(),
existing_nullable=True,
)
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.alter_column(
"created_at",
existing_type=sa.DateTime(timezone=True),
type_=postgresql.TIMESTAMP(),
existing_nullable=False,
)
with op.batch_alter_table("recording", schema=None) as batch_op:
batch_op.alter_column(
"recorded_at",
existing_type=sa.DateTime(timezone=True),
type_=postgresql.TIMESTAMP(),
existing_nullable=False,
)
with op.batch_alter_table("meeting_consent", schema=None) as batch_op:
batch_op.alter_column(
"consent_timestamp",
existing_type=sa.DateTime(timezone=True),
type_=postgresql.TIMESTAMP(),
existing_nullable=False,
)
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.alter_column(
"end_date",
existing_type=sa.DateTime(timezone=True),
type_=postgresql.TIMESTAMP(),
existing_nullable=True,
)
batch_op.alter_column(
"start_date",
existing_type=sa.DateTime(timezone=True),
type_=postgresql.TIMESTAMP(),
existing_nullable=True,
)
# ### end Alembic commands ###

View File

@@ -25,7 +25,7 @@ def upgrade() -> None:
sa.Column(
"is_shared",
sa.Boolean(),
server_default=sa.text("false"),
server_default=sa.text("0"),
nullable=False,
),
)

View File

@@ -23,10 +23,7 @@ def upgrade() -> None:
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.add_column(
sa.Column(
"is_active",
sa.Boolean(),
server_default=sa.text("true"),
nullable=False,
"is_active", sa.Boolean(), server_default=sa.text("1"), nullable=False
)
)

View File

@@ -1,41 +0,0 @@
"""add_search_optimization_indexes
Revision ID: b1c33bd09963
Revises: 9f5c78d352d6
Create Date: 2025-08-14 17:26:02.117408
"""
from typing import Sequence, Union
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "b1c33bd09963"
down_revision: Union[str, None] = "9f5c78d352d6"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# Add indexes for actual search filtering patterns used in frontend
# Based on /browse page filters: room_id and source_kind
# Index for room_id + created_at (for room-specific searches with date ordering)
op.create_index(
"idx_transcript_room_id_created_at",
"transcript",
["room_id", "created_at"],
if_not_exists=True,
)
# Index for source_kind alone (actively used filter in frontend)
op.create_index(
"idx_transcript_source_kind", "transcript", ["source_kind"], if_not_exists=True
)
def downgrade() -> None:
# Remove the indexes in reverse order
op.drop_index("idx_transcript_source_kind", "transcript", if_exists=True)
op.drop_index("idx_transcript_room_id_created_at", "transcript", if_exists=True)

View File

@@ -23,7 +23,7 @@ def upgrade() -> None:
op.add_column(
"transcript",
sa.Column(
"reviewed", sa.Boolean(), server_default=sa.text("false"), nullable=False
"reviewed", sa.Boolean(), server_default=sa.text("0"), nullable=False
),
)
# ### end Alembic commands ###

View File

@@ -1,38 +0,0 @@
"""add user table
Revision ID: bbafedfa510c
Revises: 5d6b9df9b045
Create Date: 2025-11-19 21:06:30.543262
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "bbafedfa510c"
down_revision: Union[str, None] = "5d6b9df9b045"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
op.create_table(
"user",
sa.Column("id", sa.String(), nullable=False),
sa.Column("email", sa.String(), nullable=False),
sa.Column("authentik_uid", sa.String(), nullable=False),
sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False),
sa.PrimaryKeyConstraint("id"),
)
with op.batch_alter_table("user", schema=None) as batch_op:
batch_op.create_index("idx_user_authentik_uid", ["authentik_uid"], unique=True)
batch_op.create_index("idx_user_email", ["email"], unique=False)
def downgrade() -> None:
op.drop_table("user")

View File

@@ -1,34 +0,0 @@
"""add_grace_period_fields_to_meeting
Revision ID: d4a1c446458c
Revises: 6025e9b2bef2
Create Date: 2025-08-18 18:50:37.768052
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "d4a1c446458c"
down_revision: Union[str, None] = "6025e9b2bef2"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# Add fields to track when participants left for grace period logic
op.add_column(
"meeting", sa.Column("last_participant_left_at", sa.DateTime(timezone=True))
)
op.add_column(
"meeting",
sa.Column("grace_period_minutes", sa.Integer, server_default=sa.text("15")),
)
def downgrade() -> None:
op.drop_column("meeting", "grace_period_minutes")
op.drop_column("meeting", "last_participant_left_at")

View File

@@ -1,129 +0,0 @@
"""add calendar
Revision ID: d8e204bbf615
Revises: d4a1c446458c
Create Date: 2025-09-10 19:56:22.295756
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
from sqlalchemy.dialects import postgresql
# revision identifiers, used by Alembic.
revision: str = "d8e204bbf615"
down_revision: Union[str, None] = "d4a1c446458c"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
op.create_table(
"calendar_event",
sa.Column("id", sa.String(), nullable=False),
sa.Column("room_id", sa.String(), nullable=False),
sa.Column("ics_uid", sa.Text(), nullable=False),
sa.Column("title", sa.Text(), nullable=True),
sa.Column("description", sa.Text(), nullable=True),
sa.Column("start_time", sa.DateTime(timezone=True), nullable=False),
sa.Column("end_time", sa.DateTime(timezone=True), nullable=False),
sa.Column("attendees", postgresql.JSONB(astext_type=sa.Text()), nullable=True),
sa.Column("location", sa.Text(), nullable=True),
sa.Column("ics_raw_data", sa.Text(), nullable=True),
sa.Column("last_synced", sa.DateTime(timezone=True), nullable=False),
sa.Column(
"is_deleted", sa.Boolean(), server_default=sa.text("false"), nullable=False
),
sa.Column("created_at", sa.DateTime(timezone=True), nullable=False),
sa.Column("updated_at", sa.DateTime(timezone=True), nullable=False),
sa.ForeignKeyConstraint(
["room_id"],
["room.id"],
name="fk_calendar_event_room_id",
ondelete="CASCADE",
),
sa.PrimaryKeyConstraint("id"),
sa.UniqueConstraint("room_id", "ics_uid", name="uq_room_calendar_event"),
)
with op.batch_alter_table("calendar_event", schema=None) as batch_op:
batch_op.create_index(
"idx_calendar_event_deleted",
["is_deleted"],
unique=False,
postgresql_where=sa.text("NOT is_deleted"),
)
batch_op.create_index(
"idx_calendar_event_room_start", ["room_id", "start_time"], unique=False
)
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.add_column(sa.Column("calendar_event_id", sa.String(), nullable=True))
batch_op.add_column(
sa.Column(
"calendar_metadata",
postgresql.JSONB(astext_type=sa.Text()),
nullable=True,
)
)
batch_op.create_index(
"idx_meeting_calendar_event", ["calendar_event_id"], unique=False
)
batch_op.create_foreign_key(
"fk_meeting_calendar_event_id",
"calendar_event",
["calendar_event_id"],
["id"],
ondelete="SET NULL",
)
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.add_column(sa.Column("ics_url", sa.Text(), nullable=True))
batch_op.add_column(
sa.Column(
"ics_fetch_interval", sa.Integer(), server_default="300", nullable=True
)
)
batch_op.add_column(
sa.Column(
"ics_enabled",
sa.Boolean(),
server_default=sa.text("false"),
nullable=False,
)
)
batch_op.add_column(
sa.Column("ics_last_sync", sa.DateTime(timezone=True), nullable=True)
)
batch_op.add_column(sa.Column("ics_last_etag", sa.Text(), nullable=True))
batch_op.create_index("idx_room_ics_enabled", ["ics_enabled"], unique=False)
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("room", schema=None) as batch_op:
batch_op.drop_index("idx_room_ics_enabled")
batch_op.drop_column("ics_last_etag")
batch_op.drop_column("ics_last_sync")
batch_op.drop_column("ics_enabled")
batch_op.drop_column("ics_fetch_interval")
batch_op.drop_column("ics_url")
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.drop_constraint("fk_meeting_calendar_event_id", type_="foreignkey")
batch_op.drop_index("idx_meeting_calendar_event")
batch_op.drop_column("calendar_metadata")
batch_op.drop_column("calendar_event_id")
with op.batch_alter_table("calendar_event", schema=None) as batch_op:
batch_op.drop_index("idx_calendar_event_room_start")
batch_op.drop_index(
"idx_calendar_event_deleted", postgresql_where=sa.text("NOT is_deleted")
)
op.drop_table("calendar_event")
# ### end Alembic commands ###

View File

@@ -1,43 +0,0 @@
"""remove_grace_period_fields
Revision ID: dc035ff72fd5
Revises: d8e204bbf615
Create Date: 2025-09-11 10:36:45.197588
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "dc035ff72fd5"
down_revision: Union[str, None] = "d8e204bbf615"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# Remove grace period columns from meeting table
op.drop_column("meeting", "last_participant_left_at")
op.drop_column("meeting", "grace_period_minutes")
def downgrade() -> None:
# Add back grace period columns to meeting table
op.add_column(
"meeting",
sa.Column(
"last_participant_left_at", sa.DateTime(timezone=True), nullable=True
),
)
op.add_column(
"meeting",
sa.Column(
"grace_period_minutes",
sa.Integer(),
server_default=sa.text("15"),
nullable=True,
),
)

View File

@@ -1,34 +0,0 @@
"""make meeting room_id nullable but keep foreign key
Revision ID: def1b5867d4c
Revises: 0ce521cda2ee
Create Date: 2025-09-11 09:42:18.697264
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "def1b5867d4c"
down_revision: Union[str, None] = "0ce521cda2ee"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.alter_column("room_id", existing_type=sa.VARCHAR(), nullable=True)
# ### end Alembic commands ###
def downgrade() -> None:
# ### commands auto generated by Alembic - please adjust! ###
with op.batch_alter_table("meeting", schema=None) as batch_op:
batch_op.alter_column("room_id", existing_type=sa.VARCHAR(), nullable=False)
# ### end Alembic commands ###

View File

@@ -1,28 +0,0 @@
"""add_track_keys
Revision ID: f8294b31f022
Revises: 1e49625677e4
Create Date: 2025-10-27 18:52:17.589167
"""
from typing import Sequence, Union
import sqlalchemy as sa
from alembic import op
# revision identifiers, used by Alembic.
revision: str = "f8294b31f022"
down_revision: Union[str, None] = "1e49625677e4"
branch_labels: Union[str, Sequence[str], None] = None
depends_on: Union[str, Sequence[str], None] = None
def upgrade() -> None:
with op.batch_alter_table("recording", schema=None) as batch_op:
batch_op.add_column(sa.Column("track_keys", sa.JSON(), nullable=True))
def downgrade() -> None:
with op.batch_alter_table("recording", schema=None) as batch_op:
batch_op.drop_column("track_keys")

View File

@@ -12,6 +12,7 @@ dependencies = [
"requests>=2.31.0",
"aiortc>=1.5.0",
"sortedcontainers>=2.4.0",
"loguru>=0.7.0",
"pydantic-settings>=2.0.2",
"structlog>=23.1.0",
"uvicorn[standard]>=0.23.1",
@@ -26,19 +27,19 @@ dependencies = [
"prometheus-fastapi-instrumentator>=6.1.0",
"sentencepiece>=0.1.99",
"protobuf>=4.24.3",
"profanityfilter>=2.0.6",
"celery>=5.3.4",
"redis>=5.0.1",
"python-jose[cryptography]>=3.3.0",
"python-multipart>=0.0.6",
"faster-whisper>=0.10.0",
"transformers>=4.36.2",
"black==24.1.1",
"jsonschema>=4.23.0",
"openai>=1.59.7",
"psycopg2-binary>=2.9.10",
"llama-index>=0.12.52",
"llama-index-llms-openai-like>=0.4.0",
"pytest-env>=1.1.5",
"webvtt-py>=0.5.0",
"icalendar>=6.0.0",
]
[dependency-groups]
@@ -55,9 +56,6 @@ tests = [
"httpx-ws>=0.4.1",
"pytest-httpx>=0.23.1",
"pytest-celery>=0.0.0",
"pytest-recording>=0.13.4",
"pytest-docker>=3.2.3",
"asgi-lifespan>=2.1.0",
]
aws = ["aioboto3>=11.2.0"]
evaluation = [
@@ -66,15 +64,6 @@ evaluation = [
"tqdm>=4.66.0",
"pydantic>=2.1.1",
]
local = [
"pyannote-audio>=3.3.2",
"faster-whisper>=0.10.0",
]
silero-vad = [
"silero-vad>=5.1.2",
"torch>=2.8.0",
"torchaudio>=2.8.0",
]
[tool.uv]
default-groups = [
@@ -82,21 +71,6 @@ default-groups = [
"tests",
"aws",
"evaluation",
"local",
"silero-vad"
]
[[tool.uv.index]]
name = "pytorch-cpu"
url = "https://download.pytorch.org/whl/cpu"
explicit = true
[tool.uv.sources]
torch = [
{ index = "pytorch-cpu" },
]
torchaudio = [
{ index = "pytorch-cpu" },
]
[build-system]
@@ -109,30 +83,10 @@ packages = ["reflector"]
[tool.coverage.run]
source = ["reflector"]
[tool.pytest_env]
ENVIRONMENT = "pytest"
DATABASE_URL = "postgresql://test_user:test_password@localhost:15432/reflector_test"
AUTH_BACKEND = "jwt"
[tool.pytest.ini_options]
addopts = "-ra -q --disable-pytest-warnings --cov --cov-report html -v"
testpaths = ["tests"]
asyncio_mode = "auto"
markers = [
"model_api: tests for the unified model-serving HTTP API (backend- and hardware-agnostic)",
]
[tool.ruff.lint]
select = [
"I", # isort - import sorting
"F401", # unused imports
"E402", # module level import not at top of file
"PLC0415", # import-outside-top-level - detect inline imports
]
[tool.ruff.lint.per-file-ignores]
"reflector/processors/summary/summary_builder.py" = ["E501"]
"gpu/modal_deployments/**.py" = ["PLC0415"]
"reflector/tools/**.py" = ["PLC0415"]
"migrations/versions/**.py" = ["PLC0415"]
"tests/**.py" = ["PLC0415"]

View File

@@ -27,8 +27,6 @@ from reflector.views.transcripts_upload import router as transcripts_upload_rout
from reflector.views.transcripts_webrtc import router as transcripts_webrtc_router
from reflector.views.transcripts_websocket import router as transcripts_websocket_router
from reflector.views.user import router as user_router
from reflector.views.user_api_keys import router as user_api_keys_router
from reflector.views.user_websocket import router as user_ws_router
from reflector.views.whereby import router as whereby_router
from reflector.views.zulip import router as zulip_router
@@ -68,12 +66,6 @@ app.add_middleware(
allow_headers=["*"],
)
@app.get("/health")
async def health():
return {"status": "healthy"}
# metrics
instrumentator = Instrumentator(
excluded_handlers=["/docs", "/metrics"],
@@ -93,11 +85,9 @@ app.include_router(transcripts_websocket_router, prefix="/v1")
app.include_router(transcripts_webrtc_router, prefix="/v1")
app.include_router(transcripts_process_router, prefix="/v1")
app.include_router(user_router, prefix="/v1")
app.include_router(user_api_keys_router, prefix="/v1")
app.include_router(user_ws_router, prefix="/v1")
app.include_router(zulip_router, prefix="/v1")
app.include_router(whereby_router, prefix="/v1")
app.include_router(daily_router, prefix="/v1/daily")
app.include_router(daily_router, prefix="/v1")
add_pagination(app)
# prepare celery

View File

@@ -1,33 +0,0 @@
import asyncio
import functools
from uuid import uuid4
from celery import current_task
from reflector.db import get_database
from reflector.llm import llm_session_id
def asynctask(f):
@functools.wraps(f)
def wrapper(*args, **kwargs):
async def run_with_db():
task_id = current_task.request.id if current_task else None
llm_session_id.set(task_id or f"random-{uuid4().hex}")
database = get_database()
await database.connect()
try:
return await f(*args, **kwargs)
finally:
await database.disconnect()
coro = run_with_db()
try:
loop = asyncio.get_running_loop()
except RuntimeError:
loop = None
if loop and loop.is_running():
return loop.run_until_complete(coro)
return asyncio.run(coro)
return wrapper

View File

@@ -1,18 +1,14 @@
from typing import Annotated, List, Optional
from typing import Annotated, Optional
from fastapi import Depends, HTTPException
from fastapi.security import APIKeyHeader, OAuth2PasswordBearer
from fastapi.security import OAuth2PasswordBearer
from jose import JWTError, jwt
from pydantic import BaseModel
from reflector.db.user_api_keys import user_api_keys_controller
from reflector.db.users import user_controller
from reflector.logger import logger
from reflector.settings import settings
from reflector.utils import generate_uuid4
oauth2_scheme = OAuth2PasswordBearer(tokenUrl="token", auto_error=False)
api_key_header = APIKeyHeader(name="X-API-Key", auto_error=False)
jwt_public_key = open(f"reflector/auth/jwt/keys/{settings.AUTH_JWT_PUBLIC_KEY}").read()
jwt_algorithm = settings.AUTH_JWT_ALGORITHM
@@ -30,7 +26,7 @@ class JWTException(Exception):
class UserInfo(BaseModel):
sub: str
email: Optional[str] = None
email: str
def __getitem__(self, key):
return getattr(self, key)
@@ -62,65 +58,33 @@ def authenticated(token: Annotated[str, Depends(oauth2_scheme)]):
return None
async def _authenticate_user(
jwt_token: Optional[str],
api_key: Optional[str],
jwtauth: JWTAuth,
) -> UserInfo | None:
user_infos: List[UserInfo] = []
if api_key:
user_api_key = await user_api_keys_controller.verify_key(api_key)
if user_api_key:
user_infos.append(UserInfo(sub=user_api_key.user_id, email=None))
if jwt_token:
try:
payload = jwtauth.verify_token(jwt_token)
authentik_uid = payload["sub"]
email = payload["email"]
user = await user_controller.get_by_authentik_uid(authentik_uid)
if not user:
logger.info(
f"Creating new user on first login: {authentik_uid} ({email})"
)
user = await user_controller.create_or_update(
id=generate_uuid4(),
authentik_uid=authentik_uid,
email=email,
)
user_infos.append(UserInfo(sub=user.id, email=email))
except JWTError as e:
logger.error(f"JWT error: {e}")
raise HTTPException(status_code=401, detail="Invalid authentication")
if len(user_infos) == 0:
return None
if len(set([x.sub for x in user_infos])) > 1:
raise JWTException(
status_code=401,
detail="Invalid authentication: more than one user provided",
)
return user_infos[0]
async def current_user(
jwt_token: Annotated[Optional[str], Depends(oauth2_scheme)],
api_key: Annotated[Optional[str], Depends(api_key_header)],
def current_user(
token: Annotated[Optional[str], Depends(oauth2_scheme)],
jwtauth: JWTAuth = Depends(),
):
user = await _authenticate_user(jwt_token, api_key, jwtauth)
if user is None:
if token is None:
raise HTTPException(status_code=401, detail="Not authenticated")
return user
try:
payload = jwtauth.verify_token(token)
sub = payload["sub"]
return UserInfo(sub=sub)
except JWTError as e:
logger.error(f"JWT error: {e}")
raise HTTPException(status_code=401, detail="Invalid authentication")
async def current_user_optional(
jwt_token: Annotated[Optional[str], Depends(oauth2_scheme)],
api_key: Annotated[Optional[str], Depends(api_key_header)],
def current_user_optional(
token: Annotated[Optional[str], Depends(oauth2_scheme)],
jwtauth: JWTAuth = Depends(),
):
return await _authenticate_user(jwt_token, api_key, jwtauth)
# we accept no token, but if one is provided, it must be a valid one.
if token is None:
return None
try:
payload = jwtauth.verify_token(token)
sub = payload["sub"]
email = payload["email"]
return UserInfo(sub=sub, email=email)
except JWTError as e:
logger.error(f"JWT error: {e}")
raise HTTPException(status_code=401, detail="Invalid authentication")

View File

@@ -1,6 +0,0 @@
anything about Daily.co api interaction
- webhook event shapes
- REST api client
No REST api client existing found in the wild; the official lib is about working with videocall as a bot

View File

@@ -1,110 +0,0 @@
"""
Daily.co API Module
"""
# Client
from .client import DailyApiClient, DailyApiError
# Request models
from .requests import (
CreateMeetingTokenRequest,
CreateRoomRequest,
CreateWebhookRequest,
MeetingTokenProperties,
RecordingsBucketConfig,
RoomProperties,
UpdateWebhookRequest,
)
# Response models
from .responses import (
FinishedRecordingResponse,
MeetingParticipant,
MeetingParticipantsResponse,
MeetingResponse,
MeetingTokenResponse,
RecordingResponse,
RecordingS3Info,
RoomPresenceParticipant,
RoomPresenceResponse,
RoomResponse,
WebhookResponse,
)
# Webhook utilities
from .webhook_utils import (
extract_room_name,
parse_participant_joined,
parse_participant_left,
parse_recording_error,
parse_recording_ready,
parse_recording_started,
parse_webhook_payload,
verify_webhook_signature,
)
# Webhook models
from .webhooks import (
DailyTrack,
DailyWebhookEvent,
DailyWebhookEventUnion,
ParticipantJoinedEvent,
ParticipantJoinedPayload,
ParticipantLeftEvent,
ParticipantLeftPayload,
RecordingErrorEvent,
RecordingErrorPayload,
RecordingReadyEvent,
RecordingReadyToDownloadPayload,
RecordingStartedEvent,
RecordingStartedPayload,
)
__all__ = [
# Client
"DailyApiClient",
"DailyApiError",
# Requests
"CreateRoomRequest",
"RoomProperties",
"RecordingsBucketConfig",
"CreateMeetingTokenRequest",
"MeetingTokenProperties",
"CreateWebhookRequest",
"UpdateWebhookRequest",
# Responses
"RoomResponse",
"RoomPresenceResponse",
"RoomPresenceParticipant",
"MeetingParticipantsResponse",
"MeetingParticipant",
"MeetingResponse",
"RecordingResponse",
"FinishedRecordingResponse",
"RecordingS3Info",
"MeetingTokenResponse",
"WebhookResponse",
# Webhooks
"DailyWebhookEvent",
"DailyWebhookEventUnion",
"DailyTrack",
"ParticipantJoinedEvent",
"ParticipantJoinedPayload",
"ParticipantLeftEvent",
"ParticipantLeftPayload",
"RecordingStartedEvent",
"RecordingStartedPayload",
"RecordingReadyEvent",
"RecordingReadyToDownloadPayload",
"RecordingErrorEvent",
"RecordingErrorPayload",
# Webhook utilities
"verify_webhook_signature",
"extract_room_name",
"parse_webhook_payload",
"parse_participant_joined",
"parse_participant_left",
"parse_recording_started",
"parse_recording_ready",
"parse_recording_error",
]

View File

@@ -1,573 +0,0 @@
"""
Daily.co API Client
Complete async client for Daily.co REST API with Pydantic models.
Reference: https://docs.daily.co/reference/rest-api
"""
from http import HTTPStatus
from typing import Any
import httpx
import structlog
from reflector.utils.string import NonEmptyString
from .requests import (
CreateMeetingTokenRequest,
CreateRoomRequest,
CreateWebhookRequest,
UpdateWebhookRequest,
)
from .responses import (
MeetingParticipantsResponse,
MeetingResponse,
MeetingTokenResponse,
RecordingResponse,
RoomPresenceResponse,
RoomResponse,
WebhookResponse,
)
logger = structlog.get_logger(__name__)
class DailyApiError(Exception):
"""Daily.co API error with full request/response context."""
def __init__(self, operation: str, response: httpx.Response):
self.operation = operation
self.response = response
self.status_code = response.status_code
self.response_body = response.text
self.url = str(response.url)
self.request_body = (
response.request.content.decode() if response.request.content else None
)
super().__init__(
f"Daily.co API error: {operation} failed with status {self.status_code}"
)
class DailyApiClient:
"""
Complete async client for Daily.co REST API.
Usage:
# Direct usage
client = DailyApiClient(api_key="your_api_key")
room = await client.create_room(CreateRoomRequest(name="my-room"))
await client.close() # Clean up when done
# Context manager (recommended)
async with DailyApiClient(api_key="your_api_key") as client:
room = await client.create_room(CreateRoomRequest(name="my-room"))
"""
BASE_URL = "https://api.daily.co/v1"
DEFAULT_TIMEOUT = 10.0
def __init__(
self,
api_key: NonEmptyString,
webhook_secret: NonEmptyString | None = None,
timeout: float = DEFAULT_TIMEOUT,
base_url: NonEmptyString | None = None,
):
"""
Initialize Daily.co API client.
Args:
api_key: Daily.co API key (Bearer token)
webhook_secret: Base64-encoded HMAC secret for webhook verification.
Must match the 'hmac' value provided when creating webhooks.
Generate with: base64.b64encode(os.urandom(32)).decode()
timeout: Default request timeout in seconds
base_url: Override base URL (for testing)
"""
self.api_key = api_key
self.webhook_secret = webhook_secret
self.timeout = timeout
self.base_url = base_url or self.BASE_URL
self.headers = {
"Authorization": f"Bearer {api_key}",
"Content-Type": "application/json",
}
self._client: httpx.AsyncClient | None = None
async def __aenter__(self):
return self
async def __aexit__(self, exc_type, exc_val, exc_tb):
await self.close()
async def _get_client(self) -> httpx.AsyncClient:
if self._client is None:
self._client = httpx.AsyncClient(timeout=self.timeout)
return self._client
async def close(self):
if self._client is not None:
await self._client.aclose()
self._client = None
async def _handle_response(
self, response: httpx.Response, operation: str
) -> dict[str, Any]:
"""
Handle API response with error logging.
Args:
response: HTTP response
operation: Operation name for logging (e.g., "create_room")
Returns:
Parsed JSON response
Raises:
DailyApiError: If request failed with full context
"""
if response.status_code >= 400:
logger.error(
f"Daily.co API error: {operation}",
status_code=response.status_code,
response_body=response.text,
request_body=response.request.content.decode()
if response.request.content
else None,
url=str(response.url),
)
raise DailyApiError(operation, response)
return response.json()
# ============================================================================
# ROOMS
# ============================================================================
async def create_room(self, request: CreateRoomRequest) -> RoomResponse:
"""
Create a new Daily.co room.
Reference: https://docs.daily.co/reference/rest-api/rooms/create-room
Args:
request: Room creation request with name, privacy, and properties
Returns:
Created room data including URL and ID
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.post(
f"{self.base_url}/rooms",
headers=self.headers,
json=request.model_dump(exclude_none=True),
)
data = await self._handle_response(response, "create_room")
return RoomResponse(**data)
async def get_room(self, room_name: NonEmptyString) -> RoomResponse:
"""
Get room configuration.
Args:
room_name: Daily.co room name
Returns:
Room configuration data
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.get(
f"{self.base_url}/rooms/{room_name}",
headers=self.headers,
)
data = await self._handle_response(response, "get_room")
return RoomResponse(**data)
async def get_room_presence(
self, room_name: NonEmptyString
) -> RoomPresenceResponse:
"""
Get current participants in a room (real-time presence).
Reference: https://docs.daily.co/reference/rest-api/rooms/get-room-presence
Args:
room_name: Daily.co room name
Returns:
List of currently present participants with join time and duration
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.get(
f"{self.base_url}/rooms/{room_name}/presence",
headers=self.headers,
)
data = await self._handle_response(response, "get_room_presence")
return RoomPresenceResponse(**data)
async def delete_room(self, room_name: NonEmptyString) -> None:
"""
Delete a room (idempotent - succeeds even if room doesn't exist).
Reference: https://docs.daily.co/reference/rest-api/rooms/delete-room
Args:
room_name: Daily.co room name
Raises:
httpx.HTTPStatusError: If API request fails (except 404)
"""
client = await self._get_client()
response = await client.delete(
f"{self.base_url}/rooms/{room_name}",
headers=self.headers,
)
# Idempotent delete - 404 means already deleted
if response.status_code == HTTPStatus.NOT_FOUND:
logger.debug("Room not found (already deleted)", room_name=room_name)
return
await self._handle_response(response, "delete_room")
# ============================================================================
# MEETINGS
# ============================================================================
async def get_meeting(self, meeting_id: NonEmptyString) -> MeetingResponse:
"""
Get full meeting information including participants.
Reference: https://docs.daily.co/reference/rest-api/meetings/get-meeting-information
Args:
meeting_id: Daily.co meeting/session ID
Returns:
Meeting metadata including room, duration, participants, and status
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.get(
f"{self.base_url}/meetings/{meeting_id}",
headers=self.headers,
)
data = await self._handle_response(response, "get_meeting")
return MeetingResponse(**data)
async def get_meeting_participants(
self,
meeting_id: NonEmptyString,
limit: int | None = None,
joined_after: NonEmptyString | None = None,
joined_before: NonEmptyString | None = None,
) -> MeetingParticipantsResponse:
"""
Get historical participant data from a completed meeting (paginated).
Reference: https://docs.daily.co/reference/rest-api/meetings/get-meeting-participants
Args:
meeting_id: Daily.co meeting/session ID
limit: Maximum number of participant records to return
joined_after: Return participants who joined after this participant_id
joined_before: Return participants who joined before this participant_id
Returns:
List of participants with join times and duration
Raises:
httpx.HTTPStatusError: If API request fails (404 when no more participants)
Note:
For pagination, use joined_after with the last participant_id from previous response.
Returns 404 when no more participants remain.
"""
params = {}
if limit is not None:
params["limit"] = limit
if joined_after is not None:
params["joined_after"] = joined_after
if joined_before is not None:
params["joined_before"] = joined_before
client = await self._get_client()
response = await client.get(
f"{self.base_url}/meetings/{meeting_id}/participants",
headers=self.headers,
params=params,
)
data = await self._handle_response(response, "get_meeting_participants")
return MeetingParticipantsResponse(**data)
# ============================================================================
# RECORDINGS
# ============================================================================
async def get_recording(self, recording_id: NonEmptyString) -> RecordingResponse:
"""
https://docs.daily.co/reference/rest-api/recordings/get-recording-information
Get recording metadata and status.
"""
client = await self._get_client()
response = await client.get(
f"{self.base_url}/recordings/{recording_id}",
headers=self.headers,
)
data = await self._handle_response(response, "get_recording")
return RecordingResponse(**data)
async def list_recordings(
self,
room_name: NonEmptyString | None = None,
starting_after: str | None = None,
ending_before: str | None = None,
limit: int = 100,
) -> list[RecordingResponse]:
"""
List recordings with optional filters.
Reference: https://docs.daily.co/reference/rest-api/recordings
Args:
room_name: Filter by room name
starting_after: Pagination cursor - recording ID to start after
ending_before: Pagination cursor - recording ID to end before
limit: Max results per page (default 100, max 100)
Note: starting_after/ending_before are pagination cursors (recording IDs),
NOT time filters. API returns recordings in reverse chronological order.
"""
client = await self._get_client()
params = {"limit": limit}
if room_name:
params["room_name"] = room_name
if starting_after:
params["starting_after"] = starting_after
if ending_before:
params["ending_before"] = ending_before
response = await client.get(
f"{self.base_url}/recordings",
headers=self.headers,
params=params,
)
data = await self._handle_response(response, "list_recordings")
if not isinstance(data, dict) or "data" not in data:
logger.error(
"Daily.co API returned unexpected format for list_recordings",
data_type=type(data).__name__,
data_keys=list(data.keys()) if isinstance(data, dict) else None,
data_sample=str(data)[:500],
room_name=room_name,
operation="list_recordings",
)
raise httpx.HTTPStatusError(
message=f"Unexpected response format from list_recordings: {type(data).__name__}",
request=response.request,
response=response,
)
return [RecordingResponse(**r) for r in data["data"]]
# ============================================================================
# MEETING TOKENS
# ============================================================================
async def create_meeting_token(
self, request: CreateMeetingTokenRequest
) -> MeetingTokenResponse:
"""
Create a meeting token for participant authentication.
Reference: https://docs.daily.co/reference/rest-api/meeting-tokens/create-meeting-token
Args:
request: Token properties including room name, user_id, permissions
Returns:
JWT meeting token
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.post(
f"{self.base_url}/meeting-tokens",
headers=self.headers,
json=request.model_dump(exclude_none=True),
)
data = await self._handle_response(response, "create_meeting_token")
return MeetingTokenResponse(**data)
# ============================================================================
# WEBHOOKS
# ============================================================================
async def list_webhooks(self) -> list[WebhookResponse]:
"""
List all configured webhooks for this account.
Reference: https://docs.daily.co/reference/rest-api/webhooks
Returns:
List of webhook configurations
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.get(
f"{self.base_url}/webhooks",
headers=self.headers,
)
data = await self._handle_response(response, "list_webhooks")
# Daily.co returns array directly (not paginated)
if isinstance(data, list):
return [WebhookResponse(**wh) for wh in data]
# Future-proof: handle potential pagination envelope
if isinstance(data, dict) and "data" in data:
return [WebhookResponse(**wh) for wh in data["data"]]
logger.warning("Unexpected webhook list response format", data=data)
return []
async def create_webhook(self, request: CreateWebhookRequest) -> WebhookResponse:
"""
Create a new webhook subscription.
Reference: https://docs.daily.co/reference/rest-api/webhooks
Args:
request: Webhook configuration with URL, event types, and HMAC secret
Returns:
Created webhook with UUID and state
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.post(
f"{self.base_url}/webhooks",
headers=self.headers,
json=request.model_dump(exclude_none=True),
)
data = await self._handle_response(response, "create_webhook")
return WebhookResponse(**data)
async def update_webhook(
self, webhook_uuid: NonEmptyString, request: UpdateWebhookRequest
) -> WebhookResponse:
"""
Update webhook configuration.
Note: Daily.co may not support PATCH for all fields.
Common pattern is delete + recreate.
Reference: https://docs.daily.co/reference/rest-api/webhooks
Args:
webhook_uuid: Webhook UUID to update
request: Updated webhook configuration
Returns:
Updated webhook configuration
Raises:
httpx.HTTPStatusError: If API request fails
"""
client = await self._get_client()
response = await client.patch(
f"{self.base_url}/webhooks/{webhook_uuid}",
headers=self.headers,
json=request.model_dump(exclude_none=True),
)
data = await self._handle_response(response, "update_webhook")
return WebhookResponse(**data)
async def delete_webhook(self, webhook_uuid: NonEmptyString) -> None:
"""
Delete a webhook.
Reference: https://docs.daily.co/reference/rest-api/webhooks
Args:
webhook_uuid: Webhook UUID to delete
Raises:
httpx.HTTPStatusError: If webhook not found or deletion fails
"""
client = await self._get_client()
response = await client.delete(
f"{self.base_url}/webhooks/{webhook_uuid}",
headers=self.headers,
)
await self._handle_response(response, "delete_webhook")
# ============================================================================
# HELPER METHODS
# ============================================================================
async def find_webhook_by_url(self, url: NonEmptyString) -> WebhookResponse | None:
"""
Find a webhook by its URL.
Args:
url: Webhook endpoint URL to search for
Returns:
Webhook if found, None otherwise
"""
webhooks = await self.list_webhooks()
for webhook in webhooks:
if webhook.url == url:
return webhook
return None
async def find_webhooks_by_pattern(
self, pattern: NonEmptyString
) -> list[WebhookResponse]:
"""
Find webhooks matching a URL pattern (e.g., 'ngrok').
Args:
pattern: String to match in webhook URLs
Returns:
List of matching webhooks
"""
webhooks = await self.list_webhooks()
return [wh for wh in webhooks if pattern in wh.url]

View File

@@ -1,162 +0,0 @@
"""
Daily.co API Request Models
Reference: https://docs.daily.co/reference/rest-api
"""
from typing import List, Literal
from pydantic import BaseModel, Field
from reflector.utils.string import NonEmptyString
class RecordingsBucketConfig(BaseModel):
"""
S3 bucket configuration for raw-tracks recordings.
Reference: https://docs.daily.co/reference/rest-api/rooms/create-room
"""
bucket_name: NonEmptyString = Field(description="S3 bucket name")
bucket_region: NonEmptyString = Field(description="AWS region (e.g., 'us-east-1')")
assume_role_arn: NonEmptyString = Field(
description="AWS IAM role ARN that Daily.co will assume to write recordings"
)
allow_api_access: bool = Field(
default=True,
description="Whether to allow API access to recording metadata",
)
class RoomProperties(BaseModel):
"""
Room configuration properties.
"""
enable_recording: Literal["cloud", "local", "raw-tracks"] | None = Field(
default=None,
description="Recording mode: 'cloud' for mixed, 'local' for local recording, 'raw-tracks' for multitrack, None to disable",
)
enable_chat: bool = Field(default=True, description="Enable in-meeting chat")
enable_screenshare: bool = Field(default=True, description="Enable screen sharing")
enable_knocking: bool = Field(
default=False,
description="Enable knocking for private rooms (allows participants to request access)",
)
start_video_off: bool = Field(
default=False, description="Start with video off for all participants"
)
start_audio_off: bool = Field(
default=False, description="Start with audio muted for all participants"
)
exp: int | None = Field(
None, description="Room expiration timestamp (Unix epoch seconds)"
)
recordings_bucket: RecordingsBucketConfig | None = Field(
None, description="S3 bucket configuration for raw-tracks recordings"
)
class CreateRoomRequest(BaseModel):
"""
Request to create a new Daily.co room.
Reference: https://docs.daily.co/reference/rest-api/rooms/create-room
"""
name: NonEmptyString = Field(description="Room name (must be unique within domain)")
privacy: Literal["public", "private"] = Field(
default="public", description="Room privacy setting"
)
properties: RoomProperties = Field(
default_factory=RoomProperties, description="Room configuration properties"
)
class MeetingTokenProperties(BaseModel):
"""
Properties for meeting token creation.
Reference: https://docs.daily.co/reference/rest-api/meeting-tokens/create-meeting-token
"""
room_name: NonEmptyString = Field(description="Room name this token is valid for")
user_id: NonEmptyString | None = Field(
None, description="User identifier to associate with token"
)
is_owner: bool = Field(
default=False, description="Grant owner privileges to token holder"
)
start_cloud_recording: bool = Field(
default=False, description="Automatically start cloud recording on join"
)
enable_recording_ui: bool = Field(
default=True, description="Show recording controls in UI"
)
eject_at_token_exp: bool = Field(
default=False, description="Eject participant when token expires"
)
nbf: int | None = Field(
None, description="Not-before timestamp (Unix epoch seconds)"
)
exp: int | None = Field(
None, description="Expiration timestamp (Unix epoch seconds)"
)
class CreateMeetingTokenRequest(BaseModel):
"""
Request to create a meeting token for participant authentication.
Reference: https://docs.daily.co/reference/rest-api/meeting-tokens/create-meeting-token
"""
properties: MeetingTokenProperties = Field(description="Token properties")
class CreateWebhookRequest(BaseModel):
"""
Request to create a webhook subscription.
Reference: https://docs.daily.co/reference/rest-api/webhooks
"""
url: NonEmptyString = Field(description="Webhook endpoint URL (must be HTTPS)")
eventTypes: List[
Literal[
"participant.joined",
"participant.left",
"recording.started",
"recording.ready-to-download",
"recording.error",
]
] = Field(
description="Array of event types to subscribe to (only events we handle)"
)
hmac: NonEmptyString = Field(
description="Base64-encoded HMAC secret for webhook signature verification"
)
basicAuth: NonEmptyString | None = Field(
None, description="Optional basic auth credentials for webhook endpoint"
)
class UpdateWebhookRequest(BaseModel):
"""
Request to update an existing webhook.
Note: Daily.co API may not support PATCH for webhooks.
Common pattern is to delete and recreate.
Reference: https://docs.daily.co/reference/rest-api/webhooks
"""
url: NonEmptyString | None = Field(None, description="New webhook endpoint URL")
eventTypes: List[NonEmptyString] | None = Field(
None, description="New array of event types"
)
hmac: NonEmptyString | None = Field(None, description="New HMAC secret")
basicAuth: NonEmptyString | None = Field(
None, description="New basic auth credentials"
)

View File

@@ -1,217 +0,0 @@
"""
Daily.co API Response Models
"""
from typing import Any, Dict, List, Literal
from pydantic import BaseModel, Field
from reflector.dailyco_api.webhooks import DailyTrack
from reflector.utils.string import NonEmptyString
# not documented in daily; we fill it according to observations
RecordingStatus = Literal["in-progress", "finished"]
class RoomResponse(BaseModel):
"""
Response from room creation or retrieval.
Reference: https://docs.daily.co/reference/rest-api/rooms/create-room
"""
id: NonEmptyString = Field(description="Unique room identifier (UUID)")
name: NonEmptyString = Field(description="Room name used in URLs")
api_created: bool = Field(description="Whether room was created via API")
privacy: Literal["public", "private"] = Field(description="Room privacy setting")
url: NonEmptyString = Field(description="Full room URL")
created_at: NonEmptyString = Field(description="ISO 8601 creation timestamp")
config: Dict[NonEmptyString, Any] = Field(
default_factory=dict, description="Room configuration properties"
)
class RoomPresenceParticipant(BaseModel):
"""
Participant presence information in a room.
Reference: https://docs.daily.co/reference/rest-api/rooms/get-room-presence
"""
room: NonEmptyString = Field(description="Room name")
id: NonEmptyString = Field(description="Participant session ID")
userId: NonEmptyString | None = Field(None, description="User ID if provided")
userName: NonEmptyString | None = Field(None, description="User display name")
joinTime: NonEmptyString = Field(description="ISO 8601 join timestamp")
duration: int = Field(description="Duration in room (seconds)")
class RoomPresenceResponse(BaseModel):
"""
Response from room presence endpoint.
Reference: https://docs.daily.co/reference/rest-api/rooms/get-room-presence
"""
total_count: int = Field(
description="Total number of participants currently in room"
)
data: List[RoomPresenceParticipant] = Field(
default_factory=list, description="Array of participant presence data"
)
class MeetingParticipant(BaseModel):
"""
Historical participant data from a meeting.
Reference: https://docs.daily.co/reference/rest-api/meetings/get-meeting-participants
"""
user_id: NonEmptyString | None = Field(None, description="User identifier")
participant_id: NonEmptyString = Field(description="Participant session identifier")
user_name: NonEmptyString | None = Field(None, description="User display name")
join_time: int = Field(description="Join timestamp (Unix epoch seconds)")
duration: int = Field(description="Duration in meeting (seconds)")
class MeetingParticipantsResponse(BaseModel):
"""
Response from meeting participants endpoint.
Reference: https://docs.daily.co/reference/rest-api/meetings/get-meeting-participants
"""
data: List[MeetingParticipant] = Field(
default_factory=list, description="Array of participant data"
)
class MeetingResponse(BaseModel):
"""
Response from meeting information endpoint.
Reference: https://docs.daily.co/reference/rest-api/meetings/get-meeting-information
"""
id: NonEmptyString = Field(description="Meeting session identifier (UUID)")
room: NonEmptyString = Field(description="Room name where meeting occurred")
start_time: int = Field(
description="Meeting start Unix timestamp (~15s granularity)"
)
duration: int = Field(description="Total meeting duration in seconds")
ongoing: bool = Field(description="Whether meeting is currently active")
max_participants: int = Field(description="Peak concurrent participant count")
participants: List[MeetingParticipant] = Field(
default_factory=list, description="Array of participant session data"
)
class RecordingS3Info(BaseModel):
"""
S3 bucket information for a recording.
Reference: https://docs.daily.co/reference/rest-api/recordings
"""
bucket_name: NonEmptyString
bucket_region: NonEmptyString
endpoint: NonEmptyString | None = None
class RecordingResponse(BaseModel):
"""
Response from recording retrieval endpoint (network layer).
Duration may be None for recordings still being processed by Daily.
Use FinishedRecordingResponse for recordings ready for processing.
Reference: https://docs.daily.co/reference/rest-api/recordings
"""
id: NonEmptyString = Field(description="Recording identifier")
room_name: NonEmptyString = Field(description="Room where recording occurred")
start_ts: int = Field(description="Recording start timestamp (Unix epoch seconds)")
status: RecordingStatus = Field(
description="Recording status ('in-progress' or 'finished')"
)
max_participants: int | None = Field(
None, description="Maximum participants during recording (may be missing)"
)
duration: int | None = Field(
None, description="Recording duration in seconds (None if still processing)"
)
share_token: NonEmptyString | None = Field(
None, description="Token for sharing recording"
)
s3: RecordingS3Info | None = Field(None, description="S3 bucket information")
tracks: list[DailyTrack] = Field(
default_factory=list,
description="Track list for raw-tracks recordings (always array, never null)",
)
# this is not a mistake but a deliberate Daily.co naming decision
mtgSessionId: NonEmptyString | None = Field(
None, description="Meeting session identifier (may be missing)"
)
def to_finished(self) -> "FinishedRecordingResponse | None":
"""Convert to FinishedRecordingResponse if duration is available and status is finished."""
if self.duration is None or self.status != "finished":
return None
return FinishedRecordingResponse(**self.model_dump())
class FinishedRecordingResponse(RecordingResponse):
"""
Recording with confirmed duration - ready for processing.
This model guarantees duration is present and status is finished.
"""
status: Literal["finished"] = Field(
description="Recording status (always 'finished')"
)
duration: int = Field(description="Recording duration in seconds")
class MeetingTokenResponse(BaseModel):
"""
Response from meeting token creation.
Reference: https://docs.daily.co/reference/rest-api/meeting-tokens/create-meeting-token
"""
token: NonEmptyString = Field(
description="JWT meeting token for participant authentication"
)
class WebhookResponse(BaseModel):
"""
Response from webhook creation or retrieval.
Reference: https://docs.daily.co/reference/rest-api/webhooks
"""
uuid: NonEmptyString = Field(description="Unique webhook identifier")
url: NonEmptyString = Field(description="Webhook endpoint URL")
hmac: NonEmptyString | None = Field(
None, description="Base64-encoded HMAC secret for signature verification"
)
basicAuth: NonEmptyString | None = Field(
None, description="Basic auth credentials if configured"
)
eventTypes: List[NonEmptyString] = Field(
default_factory=list,
description="Array of event types (e.g., ['recording.started', 'participant.joined'])",
)
state: Literal["ACTIVE", "FAILED"] = Field(
description="Webhook state - FAILED after 3+ consecutive failures"
)
failedCount: int = Field(default=0, description="Number of consecutive failures")
lastMomentPushed: NonEmptyString | None = Field(
None, description="ISO 8601 timestamp of last successful push"
)
domainId: NonEmptyString = Field(description="Daily.co domain/account identifier")
createdAt: NonEmptyString = Field(description="ISO 8601 creation timestamp")
updatedAt: NonEmptyString = Field(description="ISO 8601 last update timestamp")

View File

@@ -1,228 +0,0 @@
"""
Daily.co Webhook Utilities
Utilities for verifying and parsing Daily.co webhook events.
Reference: https://docs.daily.co/reference/rest-api/webhooks
"""
import base64
import hmac
from hashlib import sha256
import structlog
from .webhooks import (
DailyWebhookEvent,
ParticipantJoinedPayload,
ParticipantLeftPayload,
RecordingErrorPayload,
RecordingReadyToDownloadPayload,
RecordingStartedPayload,
)
logger = structlog.get_logger(__name__)
def verify_webhook_signature(
body: bytes,
signature: str,
timestamp: str,
webhook_secret: str,
) -> bool:
"""
Verify Daily.co webhook signature using HMAC-SHA256.
Daily.co signature verification:
1. Base64-decode the webhook secret
2. Create signed content: timestamp + '.' + body
3. Compute HMAC-SHA256(secret, signed_content)
4. Base64-encode the result
5. Compare with provided signature using constant-time comparison
Reference: https://docs.daily.co/reference/rest-api/webhooks
Args:
body: Raw request body bytes
signature: X-Webhook-Signature header value
timestamp: X-Webhook-Timestamp header value
webhook_secret: Base64-encoded HMAC secret
Returns:
True if signature is valid, False otherwise
Example:
>>> body = b'{"version":"1.0.0","type":"participant.joined",...}'
>>> signature = "abc123..."
>>> timestamp = "1234567890"
>>> secret = "your-base64-secret"
>>> is_valid = verify_webhook_signature(body, signature, timestamp, secret)
"""
if not signature or not timestamp or not webhook_secret:
logger.warning(
"Missing required data for webhook verification",
has_signature=bool(signature),
has_timestamp=bool(timestamp),
has_secret=bool(webhook_secret),
)
return False
try:
secret_bytes = base64.b64decode(webhook_secret)
signed_content = timestamp.encode() + b"." + body
expected = hmac.new(secret_bytes, signed_content, sha256).digest()
expected_b64 = base64.b64encode(expected).decode()
# Constant-time comparison to prevent timing attacks
return hmac.compare_digest(expected_b64, signature)
except (base64.binascii.Error, ValueError, TypeError, UnicodeDecodeError) as e:
logger.error(
"Webhook signature verification failed",
error=str(e),
error_type=type(e).__name__,
)
return False
def extract_room_name(event: DailyWebhookEvent) -> str | None:
"""
Extract room name from Daily.co webhook event payload.
Args:
event: Parsed webhook event
Returns:
Room name if present and is a string, None otherwise
Example:
>>> event = DailyWebhookEvent(**webhook_payload)
>>> room_name = extract_room_name(event)
"""
room = event.payload.get("room_name")
# Ensure we return a string, not any falsy value that might be in payload
return room if isinstance(room, str) else None
def parse_participant_joined(event: DailyWebhookEvent) -> ParticipantJoinedPayload:
"""
Parse participant.joined webhook event payload.
Args:
event: Webhook event with type "participant.joined"
Returns:
Parsed participant joined payload
Raises:
pydantic.ValidationError: If payload doesn't match expected schema
"""
return ParticipantJoinedPayload(**event.payload)
def parse_participant_left(event: DailyWebhookEvent) -> ParticipantLeftPayload:
"""
Parse participant.left webhook event payload.
Args:
event: Webhook event with type "participant.left"
Returns:
Parsed participant left payload
Raises:
pydantic.ValidationError: If payload doesn't match expected schema
"""
return ParticipantLeftPayload(**event.payload)
def parse_recording_started(event: DailyWebhookEvent) -> RecordingStartedPayload:
"""
Parse recording.started webhook event payload.
Args:
event: Webhook event with type "recording.started"
Returns:
Parsed recording started payload
Raises:
pydantic.ValidationError: If payload doesn't match expected schema
"""
return RecordingStartedPayload(**event.payload)
def parse_recording_ready(
event: DailyWebhookEvent,
) -> RecordingReadyToDownloadPayload:
"""
Parse recording.ready-to-download webhook event payload.
This event is sent when raw-tracks recordings are complete and uploaded to S3.
The payload includes a 'tracks' array with individual audio/video files.
Args:
event: Webhook event with type "recording.ready-to-download"
Returns:
Parsed recording ready payload with tracks array
Raises:
pydantic.ValidationError: If payload doesn't match expected schema
Example:
>>> event = DailyWebhookEvent(**webhook_payload)
>>> if event.type == "recording.ready-to-download":
... payload = parse_recording_ready(event)
... audio_tracks = [t for t in payload.tracks if t.type == "audio"]
"""
return RecordingReadyToDownloadPayload(**event.payload)
def parse_recording_error(event: DailyWebhookEvent) -> RecordingErrorPayload:
"""
Parse recording.error webhook event payload.
Args:
event: Webhook event with type "recording.error"
Returns:
Parsed recording error payload
Raises:
pydantic.ValidationError: If payload doesn't match expected schema
"""
return RecordingErrorPayload(**event.payload)
WEBHOOK_PARSERS = {
"participant.joined": parse_participant_joined,
"participant.left": parse_participant_left,
"recording.started": parse_recording_started,
"recording.ready-to-download": parse_recording_ready,
"recording.error": parse_recording_error,
}
def parse_webhook_payload(event: DailyWebhookEvent):
"""
Parse webhook event payload based on event type.
Args:
event: Webhook event
Returns:
Typed payload model based on event type, or raw dict if unknown
Example:
>>> event = DailyWebhookEvent(**webhook_payload)
>>> payload = parse_webhook_payload(event)
>>> if isinstance(payload, ParticipantJoinedPayload):
... print(f"User {payload.user_name} joined")
"""
parser = WEBHOOK_PARSERS.get(event.type)
if parser:
return parser(event)
else:
logger.warning("Unknown webhook event type", event_type=event.type)
return event.payload

View File

@@ -1,271 +0,0 @@
"""
Daily.co Webhook Event Models
Reference: https://docs.daily.co/reference/rest-api/webhooks
"""
from typing import Annotated, Any, Dict, Literal, Union
from pydantic import BaseModel, Field, field_validator
from reflector.utils.string import NonEmptyString
def normalize_timestamp_to_int(v):
"""
Normalize float timestamps to int by truncating decimal part.
Daily.co sometimes sends timestamps as floats (e.g., 1708972279.96).
Pydantic expects int for fields typed as `int`.
"""
if v is None:
return v
if isinstance(v, float):
return int(v)
return v
WebhookEventType = Literal[
"participant.joined",
"participant.left",
"recording.started",
"recording.ready-to-download",
"recording.error",
]
class DailyTrack(BaseModel):
"""
Individual audio or video track from a multitrack recording.
Reference: https://docs.daily.co/reference/rest-api/recordings
"""
type: Literal["audio", "video"]
s3Key: NonEmptyString = Field(description="S3 object key for the track file")
size: int = Field(description="File size in bytes")
class DailyWebhookEvent(BaseModel):
"""
Base structure for all Daily.co webhook events.
All events share five common fields documented below.
Reference: https://docs.daily.co/reference/rest-api/webhooks
"""
version: NonEmptyString = Field(
description="Represents the version of the event. This uses semantic versioning to inform a consumer if the payload has introduced any breaking changes"
)
type: WebhookEventType = Field(
description="Represents the type of the event described in the payload"
)
id: NonEmptyString = Field(
description="An identifier representing this specific event"
)
payload: Dict[NonEmptyString, Any] = Field(
description="An object representing the event, whose fields are described in the corresponding payload class"
)
event_ts: int = Field(
description="Documenting when the webhook itself was sent. This timestamp is different than the time of the event the webhook describes. For example, a recording.started event will contain a start_ts timestamp of when the actual recording started, and a slightly later event_ts timestamp indicating when the webhook event was sent"
)
_normalize_event_ts = field_validator("event_ts", mode="before")(
normalize_timestamp_to_int
)
class ParticipantJoinedPayload(BaseModel):
"""
Payload for participant.joined webhook event.
Reference: https://docs.daily.co/reference/rest-api/webhooks/events/participant-joined
"""
room_name: NonEmptyString | None = Field(None, description="Daily.co room name")
session_id: NonEmptyString = Field(description="Daily.co session identifier")
user_id: NonEmptyString = Field(description="User identifier (may be encoded)")
user_name: NonEmptyString | None = Field(None, description="User display name")
joined_at: int = Field(description="Join timestamp in Unix epoch seconds")
_normalize_joined_at = field_validator("joined_at", mode="before")(
normalize_timestamp_to_int
)
class ParticipantLeftPayload(BaseModel):
"""
Payload for participant.left webhook event.
Reference: https://docs.daily.co/reference/rest-api/webhooks/events/participant-left
"""
room_name: NonEmptyString | None = Field(None, description="Daily.co room name")
session_id: NonEmptyString = Field(description="Daily.co session identifier")
user_id: NonEmptyString = Field(description="User identifier (may be encoded)")
user_name: NonEmptyString | None = Field(None, description="User display name")
joined_at: int = Field(description="Join timestamp in Unix epoch seconds")
duration: int | None = Field(
None, description="Duration of participation in seconds"
)
_normalize_joined_at = field_validator("joined_at", mode="before")(
normalize_timestamp_to_int
)
class RecordingStartedPayload(BaseModel):
"""
Payload for recording.started webhook event.
Reference: https://docs.daily.co/reference/rest-api/webhooks/events/recording-started
"""
room_name: NonEmptyString | None = Field(None, description="Daily.co room name")
recording_id: NonEmptyString = Field(description="Recording identifier")
start_ts: int | None = Field(None, description="Recording start timestamp")
_normalize_start_ts = field_validator("start_ts", mode="before")(
normalize_timestamp_to_int
)
class RecordingReadyToDownloadPayload(BaseModel):
"""
Payload for recording.ready-to-download webhook event.
This is sent when raw-tracks recordings are complete and uploaded to S3.
Reference: https://docs.daily.co/reference/rest-api/webhooks/events/recording-ready-to-download
"""
type: Literal["cloud", "raw-tracks"] = Field(
description="The type of recording that was generated"
)
recording_id: NonEmptyString = Field(
description="An ID identifying the recording that was generated"
)
room_name: NonEmptyString = Field(
description="The name of the room where the recording was made"
)
start_ts: int = Field(
description="The Unix epoch time in seconds representing when the recording started"
)
status: Literal["finished"] = Field(
description="The status of the given recording (always 'finished' in ready-to-download webhook, see RecordingStatus in responses.py for full API statuses)"
)
max_participants: int = Field(
description="The number of participants on the call that were recorded"
)
duration: int = Field(description="The duration in seconds of the call")
s3_key: NonEmptyString = Field(
description="The location of the recording in the provided S3 bucket"
)
share_token: NonEmptyString | None = Field(
None, description="undocumented documented secret field"
)
tracks: list[DailyTrack] | None = Field(
None,
description="If the recording is a raw-tracks recording, a tracks field will be provided. If role permissions have been removed, the tracks field may be null",
)
_normalize_start_ts = field_validator("start_ts", mode="before")(
normalize_timestamp_to_int
)
class RecordingErrorPayload(BaseModel):
"""
Payload for recording.error webhook event.
Reference: https://docs.daily.co/reference/rest-api/webhooks/events/recording-error
"""
action: Literal["clourd-recording-err", "cloud-recording-error"] = Field(
description="A string describing the event that was emitted (both variants are documented)"
)
error_msg: NonEmptyString = Field(description="The error message returned")
instance_id: NonEmptyString = Field(
description="The recording instance ID that was passed into the start recording command"
)
room_name: NonEmptyString = Field(
description="The name of the room where the recording was made"
)
timestamp: int = Field(
description="The Unix epoch time in seconds representing when the error was emitted"
)
_normalize_timestamp = field_validator("timestamp", mode="before")(
normalize_timestamp_to_int
)
class ParticipantJoinedEvent(BaseModel):
version: NonEmptyString
type: Literal["participant.joined"]
id: NonEmptyString
payload: ParticipantJoinedPayload
event_ts: int
_normalize_event_ts = field_validator("event_ts", mode="before")(
normalize_timestamp_to_int
)
class ParticipantLeftEvent(BaseModel):
version: NonEmptyString
type: Literal["participant.left"]
id: NonEmptyString
payload: ParticipantLeftPayload
event_ts: int
_normalize_event_ts = field_validator("event_ts", mode="before")(
normalize_timestamp_to_int
)
class RecordingStartedEvent(BaseModel):
version: NonEmptyString
type: Literal["recording.started"]
id: NonEmptyString
payload: RecordingStartedPayload
event_ts: int
_normalize_event_ts = field_validator("event_ts", mode="before")(
normalize_timestamp_to_int
)
class RecordingReadyEvent(BaseModel):
version: NonEmptyString
type: Literal["recording.ready-to-download"]
id: NonEmptyString
payload: RecordingReadyToDownloadPayload
event_ts: int
_normalize_event_ts = field_validator("event_ts", mode="before")(
normalize_timestamp_to_int
)
class RecordingErrorEvent(BaseModel):
version: NonEmptyString
type: Literal["recording.error"]
id: NonEmptyString
payload: RecordingErrorPayload
event_ts: int
_normalize_event_ts = field_validator("event_ts", mode="before")(
normalize_timestamp_to_int
)
DailyWebhookEventUnion = Annotated[
Union[
ParticipantJoinedEvent,
ParticipantLeftEvent,
RecordingStartedEvent,
RecordingReadyEvent,
RecordingErrorEvent,
],
Field(discriminator="type"),
]

View File

@@ -1,51 +1,29 @@
import contextvars
from typing import Optional
import databases
import sqlalchemy
from reflector.events import subscribers_shutdown, subscribers_startup
from reflector.settings import settings
database = databases.Database(settings.DATABASE_URL)
metadata = sqlalchemy.MetaData()
_database_context: contextvars.ContextVar[Optional[databases.Database]] = (
contextvars.ContextVar("database", default=None)
)
def get_database() -> databases.Database:
"""Get database instance for current asyncio context"""
db = _database_context.get()
if db is None:
db = databases.Database(settings.DATABASE_URL)
_database_context.set(db)
return db
# import models
import reflector.db.calendar_events # noqa
import reflector.db.daily_participant_sessions # noqa
import reflector.db.meetings # noqa
import reflector.db.recordings # noqa
import reflector.db.rooms # noqa
import reflector.db.transcripts # noqa
import reflector.db.user_api_keys # noqa
import reflector.db.users # noqa
kwargs = {}
if "postgres" not in settings.DATABASE_URL:
raise Exception("Only postgres database is supported in reflector")
if "sqlite" in settings.DATABASE_URL:
kwargs["connect_args"] = {"check_same_thread": False}
engine = sqlalchemy.create_engine(settings.DATABASE_URL, **kwargs)
@subscribers_startup.append
async def database_connect(_):
database = get_database()
await database.connect()
@subscribers_shutdown.append
async def database_disconnect(_):
database = get_database()
await database.disconnect()

Some files were not shown because too many files have changed in this diff Show More