chore(main): release 0.38.2 (#906)

* refactor(auth): consolidate PUBLIC_MODE and mutation guards into reusable helpers

* fix: fix websocket test override

CHANGELOG.md

@@ -1,5 +1,32 @@
 # Changelog
 
+## [0.38.2](https://github.com/GreyhavenHQ/reflector/compare/v0.38.1...v0.38.2) (2026-03-12)
+
+
+### Bug Fixes
+
+* add auth guards to prevent anonymous access to write endpoints in non-public mode ([#907](https://github.com/GreyhavenHQ/reflector/issues/907)) ([cf6e867](https://github.com/GreyhavenHQ/reflector/commit/cf6e867cf12c42411e5a7412f6ec44eee8351665))
+* add tests that check some of the issues are already fixed ([#905](https://github.com/GreyhavenHQ/reflector/issues/905)) ([b53c8da](https://github.com/GreyhavenHQ/reflector/commit/b53c8da3981c394bdab08504b45d25f62c35495a))
+
+## [0.38.1](https://github.com/GreyhavenHQ/reflector/compare/v0.38.0...v0.38.1) (2026-03-06)
+
+
+### Bug Fixes
+
+* pin hatchet sdk version ([#903](https://github.com/GreyhavenHQ/reflector/issues/903)) ([504ca74](https://github.com/GreyhavenHQ/reflector/commit/504ca74184211eda9020d0b38ba7bd2b55d09991))
+
+## [0.38.0](https://github.com/GreyhavenHQ/reflector/compare/v0.37.0...v0.38.0) (2026-03-06)
+
+
+### Features
+
+* 3-mode selfhosted refactoring (--gpu, --cpu, --hosted) + audio token auth fallback ([#896](https://github.com/GreyhavenHQ/reflector/issues/896)) ([a682846](https://github.com/GreyhavenHQ/reflector/commit/a6828466456407c808302e9eb8dc4b4f0614dd6f))
+
+
+### Bug Fixes
+
+* improve hatchet workflow reliability ([#900](https://github.com/GreyhavenHQ/reflector/issues/900)) ([c155f66](https://github.com/GreyhavenHQ/reflector/commit/c155f669825e8e2a6e929821a1ef0bd94237dc11))
+
 ## [0.37.0](https://github.com/GreyhavenHQ/reflector/compare/v0.36.0...v0.37.0) (2026-03-03)
 
 

docker-compose.selfhosted.yml

@@ -1,11 +1,12 @@
 # Self-hosted production Docker Compose — single file for everything.
 #
-# Usage: ./scripts/setup-selfhosted.sh --gpu --ollama-gpu --garage --caddy
+# Usage: ./scripts/setup-selfhosted.sh <--gpu|--cpu|--hosted> [--ollama-gpu|--ollama-cpu] [--garage] [--caddy]
-#   or:  docker compose -f docker-compose.selfhosted.yml --profile gpu [--profile ollama-gpu] [--profile garage] [--profile caddy] up -d
+#   or:  docker compose -f docker-compose.selfhosted.yml [--profile gpu] [--profile ollama-gpu] [--profile garage] [--profile caddy] up -d
 #
-# Specialized models (pick ONE — required):
+# ML processing modes (pick ONE — required):
-#   --profile gpu          NVIDIA GPU for transcription/diarization/translation
+#   --gpu                  NVIDIA GPU container for transcription/diarization/translation (profile: gpu)
-#   --profile cpu          CPU-only for transcription/diarization/translation
+#   --cpu                  In-process CPU processing on server/worker (no ML container needed)
+#   --hosted               Remote GPU service URL (no ML container needed)
 #
 # Local LLM (optional — for summarization/topics):
 #   --profile ollama-gpu   Local Ollama with NVIDIA GPU
@@ -45,16 +46,9 @@ services:
       REDIS_HOST: redis
       CELERY_BROKER_URL: redis://redis:6379/1
       CELERY_RESULT_BACKEND: redis://redis:6379/1
-      # Specialized models via gpu/cpu container (aliased as "transcription")
+      # ML backend config comes from env_file (server/.env), set per-mode by setup script
-      TRANSCRIPT_BACKEND: modal
+      # HF_TOKEN needed for in-process pyannote diarization (--cpu mode)
-      TRANSCRIPT_URL: http://transcription:8000
+      HF_TOKEN: ${HF_TOKEN:-}
-      TRANSCRIPT_MODAL_API_KEY: selfhosted
-      DIARIZATION_BACKEND: modal
-      DIARIZATION_URL: http://transcription:8000
-      TRANSLATION_BACKEND: modal
-      TRANSLATE_URL: http://transcription:8000
-      PADDING_BACKEND: modal
-      PADDING_URL: http://transcription:8000
       # WebRTC: fixed UDP port range for ICE candidates (mapped above)
       WEBRTC_PORT_RANGE: "51000-51100"
     depends_on:
@@ -79,15 +73,8 @@ services:
       REDIS_HOST: redis
       CELERY_BROKER_URL: redis://redis:6379/1
       CELERY_RESULT_BACKEND: redis://redis:6379/1
-      TRANSCRIPT_BACKEND: modal
+      # ML backend config comes from env_file (server/.env), set per-mode by setup script
-      TRANSCRIPT_URL: http://transcription:8000
+      HF_TOKEN: ${HF_TOKEN:-}
-      TRANSCRIPT_MODAL_API_KEY: selfhosted
-      DIARIZATION_BACKEND: modal
-      DIARIZATION_URL: http://transcription:8000
-      TRANSLATION_BACKEND: modal
-      TRANSLATE_URL: http://transcription:8000
-      PADDING_BACKEND: modal
-      PADDING_URL: http://transcription:8000
     depends_on:
       postgres:
         condition: service_healthy
@@ -150,6 +137,7 @@ services:
   postgres:
     image: postgres:17-alpine
     restart: unless-stopped
+    command: ["postgres", "-c", "max_connections=200"]
     environment:
       POSTGRES_USER: reflector
       POSTGRES_PASSWORD: reflector
@@ -165,7 +153,10 @@ services:
 
   # ===========================================================
   # Specialized model containers (transcription, diarization, translation)
-  # Both gpu and cpu get alias "transcription" so server config never changes.
+  # Only the gpu profile is activated by the setup script (--gpu mode).
+  # The cpu service definition is kept for manual/standalone use but is
+  # NOT activated by --cpu mode (which uses in-process local backends).
+  # Both services get alias "transcription" so server config never changes.
   # ===========================================================
 
   gpu:

docs/create-docs.sh

@@ -254,15 +254,15 @@ Reflector can run completely offline:
 Control where each step happens:
 
 ```yaml
-# All local processing
+# All in-process processing
-TRANSCRIPT_BACKEND=local
+TRANSCRIPT_BACKEND=whisper
-DIARIZATION_BACKEND=local
+DIARIZATION_BACKEND=pyannote
-TRANSLATION_BACKEND=local
+TRANSLATION_BACKEND=marian
 
 # Hybrid approach
-TRANSCRIPT_BACKEND=modal  # Fast GPU processing
+TRANSCRIPT_BACKEND=modal    # Fast GPU processing
-DIARIZATION_BACKEND=local # Sensitive speaker data
+DIARIZATION_BACKEND=pyannote # Sensitive speaker data
-TRANSLATION_BACKEND=modal  # Non-sensitive translation
+TRANSLATION_BACKEND=modal    # Non-sensitive translation
 ```
 
 ### Storage Options

docs/package.json

@@ -57,7 +57,8 @@
       "minimatch@>=9.0.0 <9.0.7": "9.0.7",
       "lodash@<4.17.23": "4.17.23",
       "js-yaml@<4.1.1": "4.1.1",
-      "gray-matter": "github:jonschlinkert/gray-matter#234163e"
+      "gray-matter": "github:jonschlinkert/gray-matter#234163e",
+      "serialize-javascript": "7.0.4"
     }
   }
 }

docs/pnpm-lock.yaml

@@ -11,6 +11,7 @@ overrides:
   lodash@<4.17.23: 4.17.23
   js-yaml@<4.1.1: 4.1.1
   gray-matter: github:jonschlinkert/gray-matter#234163e
+  serialize-javascript: 7.0.4
 
 importers:
 
@@ -1489,36 +1490,42 @@ packages:
     engines: {node: '>= 10.0.0'}
     cpu: [arm]
     os: [linux]
+    libc: [glibc]
 
   '@parcel/watcher-linux-arm-musl@2.5.6':
     resolution: {integrity: sha512-Ve3gUCG57nuUUSyjBq/MAM0CzArtuIOxsBdQ+ftz6ho8n7s1i9E1Nmk/xmP323r2YL0SONs1EuwqBp2u1k5fxg==}
     engines: {node: '>= 10.0.0'}
     cpu: [arm]
     os: [linux]
+    libc: [musl]
 
   '@parcel/watcher-linux-arm64-glibc@2.5.6':
     resolution: {integrity: sha512-f2g/DT3NhGPdBmMWYoxixqYr3v/UXcmLOYy16Bx0TM20Tchduwr4EaCbmxh1321TABqPGDpS8D/ggOTaljijOA==}
     engines: {node: '>= 10.0.0'}
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   '@parcel/watcher-linux-arm64-musl@2.5.6':
     resolution: {integrity: sha512-qb6naMDGlbCwdhLj6hgoVKJl2odL34z2sqkC7Z6kzir8b5W65WYDpLB6R06KabvZdgoHI/zxke4b3zR0wAbDTA==}
     engines: {node: '>= 10.0.0'}
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   '@parcel/watcher-linux-x64-glibc@2.5.6':
     resolution: {integrity: sha512-kbT5wvNQlx7NaGjzPFu8nVIW1rWqV780O7ZtkjuWaPUgpv2NMFpjYERVi0UYj1msZNyCzGlaCWEtzc+exjMGbQ==}
     engines: {node: '>= 10.0.0'}
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   '@parcel/watcher-linux-x64-musl@2.5.6':
     resolution: {integrity: sha512-1JRFeC+h7RdXwldHzTsmdtYR/Ku8SylLgTU/reMuqdVD7CtLwf0VR1FqeprZ0eHQkO0vqsbvFLXUmYm/uNKJBg==}
     engines: {node: '>= 10.0.0'}
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   '@parcel/watcher-win32-arm64@2.5.6':
     resolution: {integrity: sha512-3ukyebjc6eGlw9yRt678DxVF7rjXatWiHvTXqphZLvo7aC5NdEgFufVwjFfY51ijYEWpXbqF5jtrK275z52D4Q==}
@@ -1729,10 +1736,6 @@ packages:
     resolution: {integrity: sha512-+PmQX0PiAYPMeVYe237LJAYvOMYW1j2rH5YROyS3b4CTVJum34HfRvKvAzozHAQG0TnHNdUfY9nCeUyRAs//cw==}
     engines: {node: '>=14.16'}
 
-  '@trysound/sax@0.2.0':
-    resolution: {integrity: sha512-L7z9BgrNEcYyUYtF+HaEfiS5ebkh9jXqbszz7pC0hRBPaatV0XjSD3+eHrpqFemQfgwiFF0QPIarnIihIDn7OA==}
-    engines: {node: '>=10.13.0'}
-
   '@types/body-parser@1.19.6':
     resolution: {integrity: sha512-HLFeCYgz89uk22N5Qg3dvGvsv46B8GLvKKo1zKG4NybA8U2DiEO3w9lqGg29t/tfLRJpJ6iQxnVw4OnB7MoM9g==}
 
@@ -3010,8 +3013,9 @@ packages:
     resolution: {integrity: sha512-cgwlv/1iFQiFnU96XXgROh8xTeetsnJiDsTc7TYCLFd9+/WNkIqPTxiM/8pSd8VIrhXGTf1Ny1q1hquVqDJB5w==}
     engines: {node: '>= 4'}
 
-  dompurify@3.3.1:
+  dompurify@3.3.2:
-    resolution: {integrity: sha512-qkdCKzLNtrgPFP1Vo+98FRzJnBRGe4ffyCea9IwHB1fyxPOeNTHpLKYGd4Uk9xvNoH0ZoOjwZxNptyMwqrId1Q==}
+    resolution: {integrity: sha512-6obghkliLdmKa56xdbLOpUZ43pAR6xFy1uOrxBaIDjT+yaRuuybLjGS9eVBoSR/UPU5fq3OXClEHLJNGvbxKpQ==}
+    engines: {node: '>=20'}
 
   domutils@2.8.0:
     resolution: {integrity: sha512-w96Cjofp72M5IIhpjgobBimYEfoPjx1Vx0BSX9P30WBdZW2WIKU0T1Bd0kz2eNZ9ikjKgHbEyKx8BB6H1L3h3A==}
@@ -3406,8 +3410,8 @@ packages:
   graphlib@2.1.8:
     resolution: {integrity: sha512-jcLLfkpoVGmH7/InMC/1hIvOPSUh38oJtGhvrOFGzioE1DZ+0YW16RgmOJhHiuWTvGiJQ9Z1Ik43JvkRPRvE+A==}
 
-  gray-matter@https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e:
+  gray-matter@https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e317c87fe031e9368ffabde9c9149ce3ec:
-    resolution: {tarball: https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e}
+    resolution: {tarball: https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e317c87fe031e9368ffabde9c9149ce3ec}
     version: 4.0.3
     engines: {node: '>=6.0'}
 
@@ -3599,8 +3603,8 @@ packages:
   immer@11.1.4:
     resolution: {integrity: sha512-XREFCPo6ksxVzP4E0ekD5aMdf8WMwmdNaz6vuvxgI40UaEiu6q3p8X52aU6GdyvLY3XXX/8R7JOTXStz/nBbRw==}
 
-  immutable@5.1.4:
+  immutable@5.1.5:
-    resolution: {integrity: sha512-p6u1bG3YSnINT5RQmx/yRZBpenIl30kVxkTLDyHLIMk0gict704Q9n+thfDI7lTRm9vXdDYutVzXhzcThxTnXA==}
+    resolution: {integrity: sha512-t7xcm2siw+hlUM68I+UEOK+z84RzmN59as9DZ7P1l0994DKUWV7UXBMQZVxaoMSRQ+PBZbHCOoBt7a2wxOMt+A==}
 
   import-fresh@3.3.1:
     resolution: {integrity: sha512-TR3KfrTZTYLPB6jUjfx6MF9WcWrHL9su5TObK4ZkYgBdWKPOFoSoQIdEuTuR82pmtxH2spWG9h6etwfr1pLBqQ==}
@@ -5060,9 +5064,6 @@ packages:
     resolution: {integrity: sha512-WuyALRjWPDGtt/wzJiadO5AXY+8hZ80hVpe6MyivgraREW751X3SbhRvG3eLKOYN+8VEvqLcf3wdnt44Z4S4SA==}
     engines: {node: '>=10'}
 
-  randombytes@2.1.0:
-    resolution: {integrity: sha512-vYl3iOX+4CKUWuxGi9Ukhie6fsqXqS9FE2Zaic4tNFD2N2QQaXOMFbuKK4QmDHC0JO6B1Zp41J0LpT0oR68amQ==}
-
   range-parser@1.2.0:
     resolution: {integrity: sha512-kA5WQoNVo4t9lNx2kQNFCxKeBl5IbbSNBl1M/tLkw9WCn+hxNBAW5Qh8gdhs63CJnhjJ2zQWFoqPJP2sK1AV5A==}
     engines: {node: '>= 0.6'}
@@ -5385,6 +5386,10 @@ packages:
     resolution: {integrity: sha512-1n3r/tGXO6b6VXMdFT54SHzT9ytu9yr7TaELowdYpMqY/Ao7EnlQGmAQ1+RatX7Tkkdm6hONI2owqNx2aZj5Sw==}
     engines: {node: '>=11.0.0'}
 
+  sax@1.5.0:
+    resolution: {integrity: sha512-21IYA3Q5cQf089Z6tgaUTr7lDAyzoTPx5HRtbhsME8Udispad8dC/+sziTNugOEx54ilvatQ9YCzl4KQLPcRHA==}
+    engines: {node: '>=11.0.0'}
+
   scheduler@0.27.0:
     resolution: {integrity: sha512-eNv+WrVbKu1f3vbYJT/xtiF5syA5HPIMtf9IgY/nKg0sWqzAUEvqY/xm7OcZc/qafLx/iO9FgOmeSAp4v5ti/Q==}
 
@@ -5435,8 +5440,9 @@ packages:
     resolution: {integrity: sha512-VMbMxbDeehAxpOtWJXlcUS5E8iXh6QmN+BkRX1GARS3wRaXEEgzCcB10gTQazO42tpNIya8xIyNx8fll1OFPrg==}
     engines: {node: '>= 0.8.0'}
 
-  serialize-javascript@6.0.2:
+  serialize-javascript@7.0.4:
-    resolution: {integrity: sha512-Saa1xPByTTq2gdeFZYLLo+RFE35NHZkAbqZeWNd3BpzppeVisAqpDjcp8dyf6uIvEqJRd46jemmyA4iFIeVk8g==}
+    resolution: {integrity: sha512-DuGdB+Po43Q5Jxwpzt1lhyFSYKryqoNjQSA9M92tyw0lyHIOur+XCalOUe0KTJpyqzT8+fQ5A0Jf7vCx/NKmIg==}
+    engines: {node: '>=20.0.0'}
 
   serve-handler@6.1.6:
     resolution: {integrity: sha512-x5RL9Y2p5+Sh3D38Fh9i/iQ5ZK+e4xuXRd/pGbM4D13tgo/MGwbttUk8emytcr1YYzBYs+apnUngBDFYfpjPuQ==}
@@ -5675,8 +5681,8 @@ packages:
   svg-parser@2.0.4:
     resolution: {integrity: sha512-e4hG1hRwoOdRb37cIMSgzNsxyzKfayW6VOflrwvR+/bzrkyxY/31WkbgnQpgtrNp1SdpJvpUAGTa/ZoiPNDuRQ==}
 
-  svgo@3.3.2:
+  svgo@3.3.3:
-    resolution: {integrity: sha512-OoohrmuUlBs8B8o6MB2Aevn+pRIH9zDALSR+6hhqVfa6fRwG/Qw9VUMSMW9VNg2CFc/MTIfabtdOVl9ODIJjpw==}
+    resolution: {integrity: sha512-+wn7I4p7YgJhHs38k2TNjy1vCfPIfLIJWR5MnCStsN8WuuTcBnRKcMHQLMM2ijxGZmDoZwNv8ipl5aTTen62ng==}
     engines: {node: '>=14.0.0'}
     hasBin: true
 
@@ -8156,7 +8162,7 @@ snapshots:
       fs-extra: 11.3.3
       github-slugger: 1.5.0
       globby: 11.1.0
-      gray-matter: https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e
+      gray-matter: https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e317c87fe031e9368ffabde9c9149ce3ec
       jiti: 1.21.7
       js-yaml: 4.1.1
       lodash: 4.17.23
@@ -8732,7 +8738,7 @@ snapshots:
       '@svgr/core': 8.1.0(typescript@5.9.3)
       cosmiconfig: 8.3.6(typescript@5.9.3)
       deepmerge: 4.3.1
-      svgo: 3.3.2
+      svgo: 3.3.3
     transitivePeerDependencies:
       - typescript
 
@@ -8754,8 +8760,6 @@ snapshots:
     dependencies:
       defer-to-connect: 2.0.1
 
-  '@trysound/sax@0.2.0': {}
-
   '@types/body-parser@1.19.6':
     dependencies:
       '@types/connect': 3.4.38
@@ -9670,7 +9674,7 @@ snapshots:
       globby: 13.2.2
       normalize-path: 3.0.0
       schema-utils: 4.3.3
-      serialize-javascript: 6.0.2
+      serialize-javascript: 7.0.4
       webpack: 5.105.3
 
   core-js-compat@3.48.0:
@@ -9748,7 +9752,7 @@ snapshots:
       jest-worker: 29.7.0
       postcss: 8.5.6
       schema-utils: 4.3.3
-      serialize-javascript: 6.0.2
+      serialize-javascript: 7.0.4
       webpack: 5.105.3
     optionalDependencies:
       clean-css: 5.3.3
@@ -10228,7 +10232,7 @@ snapshots:
     dependencies:
       domelementtype: 2.3.0
 
-  dompurify@3.3.1:
+  dompurify@3.3.2:
     optionalDependencies:
       '@types/trusted-types': 2.0.7
 
@@ -10654,7 +10658,7 @@ snapshots:
     dependencies:
       lodash: 4.17.23
 
-  gray-matter@https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e:
+  gray-matter@https://codeload.github.com/jonschlinkert/gray-matter/tar.gz/234163e317c87fe031e9368ffabde9c9149ce3ec:
     dependencies:
       js-yaml: 4.1.1
       kind-of: 6.0.3
@@ -10933,7 +10937,7 @@ snapshots:
 
   immer@11.1.4: {}
 
-  immutable@5.1.4: {}
+  immutable@5.1.5: {}
 
   import-fresh@3.3.1:
     dependencies:
@@ -11466,7 +11470,7 @@ snapshots:
       d3-sankey: 0.12.3
       dagre-d3-es: 7.0.13
       dayjs: 1.11.19
-      dompurify: 3.3.1
+      dompurify: 3.3.2
       katex: 0.16.33
       khroma: 2.1.0
       lodash-es: 4.17.23
@@ -12598,7 +12602,7 @@ snapshots:
     dependencies:
       postcss: 8.5.6
       postcss-value-parser: 4.2.0
-      svgo: 3.3.2
+      svgo: 3.3.3
 
   postcss-unique-selectors@6.0.4(postcss@8.5.6):
     dependencies:
@@ -12709,10 +12713,6 @@ snapshots:
 
   quick-lru@5.1.1: {}
 
-  randombytes@2.1.0:
-    dependencies:
-      safe-buffer: 5.2.1
-
   range-parser@1.2.0: {}
 
   range-parser@1.2.1: {}
@@ -13091,13 +13091,15 @@ snapshots:
   sass@1.97.3:
     dependencies:
       chokidar: 4.0.3
-      immutable: 5.1.4
+      immutable: 5.1.5
       source-map-js: 1.2.1
     optionalDependencies:
       '@parcel/watcher': 2.5.6
 
   sax@1.4.4: {}
 
+  sax@1.5.0: {}
+
   scheduler@0.27.0: {}
 
   schema-dts@1.1.5: {}
@@ -13157,9 +13159,7 @@ snapshots:
     transitivePeerDependencies:
       - supports-color
 
-  serialize-javascript@6.0.2:
+  serialize-javascript@7.0.4: {}
-    dependencies:
-      randombytes: 2.1.0
 
   serve-handler@6.1.6:
     dependencies:
@@ -13444,15 +13444,15 @@ snapshots:
 
   svg-parser@2.0.4: {}
 
-  svgo@3.3.2:
+  svgo@3.3.3:
     dependencies:
-      '@trysound/sax': 0.2.0
       commander: 7.2.0
       css-select: 5.2.2
       css-tree: 2.3.1
       css-what: 6.2.2
       csso: 5.0.5
       picocolors: 1.1.1
+      sax: 1.5.0
 
   swagger2openapi@7.0.8:
     dependencies:
@@ -13477,7 +13477,7 @@ snapshots:
       '@jridgewell/trace-mapping': 0.3.31
       jest-worker: 27.5.1
       schema-utils: 4.3.3
-      serialize-javascript: 6.0.2
+      serialize-javascript: 7.0.4
       terser: 5.46.0
       webpack: 5.105.3
 

docsv2/selfhosted-production.md

@@ -53,9 +53,12 @@ cd reflector
 # Same but without a domain (self-signed cert, access via IP):
 ./scripts/setup-selfhosted.sh --gpu --ollama-gpu --garage --caddy
 
-# CPU-only (same, but slower):
+# CPU-only (in-process ML, no GPU container):
 ./scripts/setup-selfhosted.sh --cpu --ollama-cpu --garage --caddy
 
+# Remote GPU service (your own hosted GPU, no local ML container):
+./scripts/setup-selfhosted.sh --hosted --garage --caddy
+
 # With password authentication (single admin user):
 ./scripts/setup-selfhosted.sh --gpu --ollama-gpu --garage --caddy --password mysecretpass
 
@@ -65,14 +68,15 @@ cd reflector
 
 That's it. The script generates env files, secrets, starts all containers, waits for health checks, and prints the URL.
 
-## Specialized Models (Required)
+## ML Processing Modes (Required)
 
-Pick `--gpu` or `--cpu`. This determines how **transcription, diarization, translation, and audio padding** run:
+Pick `--gpu`, `--cpu`, or `--hosted`. This determines how **transcription, diarization, translation, and audio padding** run:
 
 | Flag | What it does | Requires |
 |------|-------------|----------|
-| `--gpu` | NVIDIA GPU acceleration for ML models | NVIDIA GPU + drivers + `nvidia-container-toolkit` |
+| `--gpu` | NVIDIA GPU container for ML models | NVIDIA GPU + drivers + `nvidia-container-toolkit` |
-| `--cpu` | CPU-only (slower but works without GPU) | 8+ cores, 32GB+ RAM recommended |
+| `--cpu` | In-process CPU processing on server/worker (no ML container) | 8+ cores, 16GB+ RAM (32GB recommended for large files) |
+| `--hosted` | Remote GPU service URL (no local ML container) | A running GPU service instance (e.g. `gpu/self_hosted/`) |
 
 ## Local LLM (Optional)
 
@@ -130,9 +134,11 @@ Browse all available models at https://ollama.com/library.
 
 - **`--gpu --ollama-gpu`**: Best for servers with NVIDIA GPU. Fully self-contained, no external API keys needed.
 - **`--cpu --ollama-cpu`**: No GPU available but want everything self-contained. Slower but works.
+- **`--hosted --ollama-cpu`**: Remote GPU for ML, local CPU for LLM. Great when you have a separate GPU server.
 - **`--gpu --ollama-cpu`**: GPU for transcription, CPU for LLM. Saves GPU VRAM for ML models.
 - **`--gpu`**: Have NVIDIA GPU but prefer a cloud LLM (faster/better summaries with GPT-4, Claude, etc.).
 - **`--cpu`**: No GPU, prefer cloud LLM. Slowest transcription but best summary quality.
+- **`--hosted`**: Remote GPU, cloud LLM. No local ML at all.
 
 ## Other Optional Flags
 
@@ -160,7 +166,7 @@ Without `--caddy` or `--domain`, no ports are exposed. Point your own reverse pr
 4. **Generate `www/.env`** — Auto-detects server IP, sets URLs
 5. **Storage setup** — Either initializes Garage (bucket, keys, permissions) or prompts for external S3 credentials
 6. **Caddyfile** — Generates domain-specific (Let's Encrypt) or IP-specific (self-signed) configuration
-7. **Build & start** — Always builds GPU/CPU model image from source. With `--build`, also builds backend and frontend from source; otherwise pulls prebuilt images from the registry
+7. **Build & start** — For `--gpu`, builds the GPU model image from source. For `--cpu` and `--hosted`, no ML container is built. With `--build`, also builds backend and frontend from source; otherwise pulls prebuilt images from the registry
 8. **Auto-detects video platforms** — If `DAILY_API_KEY` is found in `server/.env`, generates `.env.hatchet` (dashboard URL/cookie config), starts Hatchet workflow engine, and generates an API token. If any video platform is configured, enables the Rooms feature
 9. **Health checks** — Waits for each service, pulls Ollama model if needed, warns about missing LLM config
 
@@ -181,7 +187,7 @@ Without `--caddy` or `--domain`, no ports are exposed. Point your own reverse pr
 | `ADMIN_PASSWORD_HASH` | PBKDF2 hash for password auth | *(unset)* |
 | `WEBRTC_HOST` | IP advertised in WebRTC ICE candidates | Auto-detected (server IP) |
 | `TRANSCRIPT_URL` | Specialized model endpoint | `http://transcription:8000` |
-| `PADDING_BACKEND` | Audio padding backend (`local` or `modal`) | `modal` (selfhosted), `local` (default) |
+| `PADDING_BACKEND` | Audio padding backend (`pyav` or `modal`) | `modal` (selfhosted), `pyav` (default) |
 | `PADDING_URL` | Audio padding endpoint (when `PADDING_BACKEND=modal`) | `http://transcription:8000` |
 | `LLM_URL` | OpenAI-compatible LLM endpoint | Auto-set for Ollama modes |
 | `LLM_API_KEY` | LLM API key | `not-needed` for Ollama |
@@ -604,10 +610,9 @@ The setup script is idempotent — it won't overwrite existing secrets or env va
           │              │            │
           v              v            v
     ┌───────────┐  ┌─────────┐  ┌─────────┐
-    │transcription│  │postgres │  │  redis  │
+    │ ML models │  │postgres │  │  redis  │
-    │(gpu/cpu)  │  │ :5432   │  │ :6379   │
+    │ (varies)  │  │ :5432   │  │ :6379   │
-    │ :8000     │  └─────────┘  └─────────┘
+    └───────────┘  └─────────┘  └─────────┘
-    └───────────┘
           │
     ┌─────┴─────┐     ┌─────────┐
     │  ollama   │     │ garage  │
@@ -622,6 +627,11 @@ The setup script is idempotent — it won't overwrite existing secrets or env va
     │  │ :8888   │──│  -cpu / -llm  │  │
     │  └─────────┘  └───────────────┘  │
     └───────────────────────────────────┘
+
+ML models box varies by mode:
+  --gpu:    Local GPU container (transcription:8000)
+  --cpu:    In-process on server/worker (no container)
+  --hosted: Remote GPU service (user URL)
 ```
 
 All services communicate over Docker's internal network. Only Caddy (if enabled) exposes ports to the internet. Hatchet services are only started when `DAILY_API_KEY` is configured.

gpu/self_hosted/pyproject.toml

@@ -11,10 +11,11 @@ dependencies = [
     "faster-whisper>=1.1.0",
     "librosa==0.10.1",
     "numpy<2",
-    "silero-vad==5.1.0",
+    "silero-vad==5.1.2",
     "transformers>=4.35.0",
     "sentencepiece",
-    "pyannote.audio==3.1.0",
+    "pyannote.audio==3.4.0",
+    "pytorch-lightning<2.6",
     "torchaudio>=2.3.0",
     "av>=13.1.0",
 ]

gpu/self_hosted/uv.lock

@@ -1742,7 +1742,7 @@ wheels = [
 
 [[package]]
 name = "pyannote-audio"
-version = "3.1.0"
+version = "3.4.0"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "asteroid-filterbanks" },
@@ -1765,9 +1765,9 @@ dependencies = [
     { name = "torchaudio" },
     { name = "torchmetrics" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/ad/55/7253267c35e2aa9188b1d86cba121eb5bdd91ed12d3194488625a008cae7/pyannote.audio-3.1.0.tar.gz", hash = "sha256:da04705443d3b74607e034d3ca88f8b572c7e9672dd9a4199cab65a0dbc33fad", size = 14812058, upload-time = "2023-11-16T12:26:38.939Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/ec/1e/efe9619c38f1281ddf21640654d8ea9e3f67c459b76f78657b26d8557bbe/pyannote_audio-3.4.0.tar.gz", hash = "sha256:d523d883cb8d37cb6daf99f3ba83f9138bb193646ad71e6eae7deb89d8ddd642", size = 804850, upload-time = "2025-09-09T07:04:51.17Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/a1/37/158859ce4c45b5ba2dca40b53b0c10d36f935b7f6d4e737298397167c8b1/pyannote.audio-3.1.0-py2.py3-none-any.whl", hash = "sha256:66ab485728c6e141760e80555cb7a083e7be824cd528cc79b9e6f7d6421a91ae", size = 208592, upload-time = "2023-11-16T12:26:36.726Z" },
+    { url = "https://files.pythonhosted.org/packages/79/13/620c6f711b723653092fd063bfee82a6af5ea3a4d3c42efc53ce623a7f4d/pyannote_audio-3.4.0-py2.py3-none-any.whl", hash = "sha256:36e38f058059f46da3478dda581cda53d9d85a21173a3e70bbdbc3ba93b5e1b7", size = 897789, upload-time = "2025-09-09T07:04:49.464Z" },
 ]
 
 [[package]]
@@ -2075,6 +2075,7 @@ dependencies = [
     { name = "librosa" },
     { name = "numpy" },
     { name = "pyannote-audio" },
+    { name = "pytorch-lightning" },
     { name = "sentencepiece" },
     { name = "silero-vad" },
     { name = "torch" },
@@ -2090,9 +2091,10 @@ requires-dist = [
     { name = "faster-whisper", specifier = ">=1.1.0" },
     { name = "librosa", specifier = "==0.10.1" },
     { name = "numpy", specifier = "<2" },
-    { name = "pyannote-audio", specifier = "==3.1.0" },
+    { name = "pyannote-audio", specifier = "==3.4.0" },
+    { name = "pytorch-lightning", specifier = "<2.6" },
     { name = "sentencepiece" },
-    { name = "silero-vad", specifier = "==5.1.0" },
+    { name = "silero-vad", specifier = "==5.1.2" },
     { name = "torch", specifier = ">=2.3.0" },
     { name = "torchaudio", specifier = ">=2.3.0" },
     { name = "transformers", specifier = ">=4.35.0" },
@@ -2472,16 +2474,16 @@ wheels = [
 
 [[package]]
 name = "silero-vad"
-version = "5.1"
+version = "5.1.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "onnxruntime" },
     { name = "torch" },
     { name = "torchaudio" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/7c/5d/b912e45d21b8b61859a552554893222d2cdebfd0f9afa7e8ba69c7a3441a/silero_vad-5.1.tar.gz", hash = "sha256:c644275ba5df06cee596cc050ba0bd1e0f5237d1abfa44d58dd4618f6e77434d", size = 3996829, upload-time = "2024-07-09T13:19:24.181Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/b1/b4/d0311b2e6220a11f8f4699f4a278cb088131573286cdfe804c87c7eb5123/silero_vad-5.1.2.tar.gz", hash = "sha256:c442971160026d2d7aa0ad83f0c7ee86c89797a65289fe625c8ea59fc6fb828d", size = 5098526, upload-time = "2024-10-09T09:50:47.019Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/0e/be/0fdbc72030b93d6f55107490d5d2185ddf0dbabdc921f589649d3e92ccd5/silero_vad-5.1-py3-none-any.whl", hash = "sha256:ecb50b484f538f7a962ce5cd3c07120d9db7b9d5a0c5861ccafe459856f22c8f", size = 3939986, upload-time = "2024-07-09T13:19:21.383Z" },
+    { url = "https://files.pythonhosted.org/packages/98/f7/5ae11d13fbb733cd3bfd7ff1c3a3902e6f55437df4b72307c1f168146268/silero_vad-5.1.2-py3-none-any.whl", hash = "sha256:93b41953d7774b165407fda6b533c119c5803864e367d5034dc626c82cfdf661", size = 5026737, upload-time = "2024-10-09T09:50:44.355Z" },
 ]
 
 [[package]]

scripts/setup-selfhosted.sh

@@ -4,11 +4,12 @@
 # Single script to configure and launch everything on one server.
 #
 # Usage:
-#   ./scripts/setup-selfhosted.sh <--gpu|--cpu> [--ollama-gpu|--ollama-cpu] [--llm-model MODEL] [--garage] [--caddy] [--domain DOMAIN] [--password PASSWORD] [--build]
+#   ./scripts/setup-selfhosted.sh <--gpu|--cpu|--hosted> [--ollama-gpu|--ollama-cpu] [--llm-model MODEL] [--garage] [--caddy] [--domain DOMAIN] [--password PASSWORD] [--build]
 #
-# Specialized models (pick ONE — required):
+# ML processing modes (pick ONE — required):
-#   --gpu              NVIDIA GPU for transcription/diarization/translation
+#   --gpu              NVIDIA GPU container for transcription/diarization/translation
-#   --cpu              CPU-only for transcription/diarization/translation (slower)
+#   --cpu              In-process CPU processing (no ML container, slower)
+#   --hosted           Remote GPU service URL (no ML container)
 #
 # Local LLM (optional — for summarization & topic detection):
 #   --ollama-gpu       Local Ollama with NVIDIA GPU acceleration
@@ -29,6 +30,7 @@
 #   ./scripts/setup-selfhosted.sh --gpu --ollama-gpu --garage --caddy
 #   ./scripts/setup-selfhosted.sh --gpu --ollama-gpu --garage --caddy --domain reflector.example.com
 #   ./scripts/setup-selfhosted.sh --cpu --ollama-cpu --garage --caddy
+#   ./scripts/setup-selfhosted.sh --hosted --garage --caddy
 #   ./scripts/setup-selfhosted.sh --gpu --ollama-gpu --llm-model mistral --garage --caddy
 #   ./scripts/setup-selfhosted.sh --gpu --garage --caddy --password mysecretpass
 #   ./scripts/setup-selfhosted.sh --gpu --garage --caddy
@@ -183,11 +185,14 @@ for i in "${!ARGS[@]}"; do
     arg="${ARGS[$i]}"
     case "$arg" in
         --gpu)
-            [[ -n "$MODEL_MODE" ]] && { err "Cannot combine --gpu and --cpu. Pick one."; exit 1; }
+            [[ -n "$MODEL_MODE" ]] && { err "Cannot combine --gpu, --cpu, and --hosted. Pick one."; exit 1; }
             MODEL_MODE="gpu" ;;
         --cpu)
-            [[ -n "$MODEL_MODE" ]] && { err "Cannot combine --gpu and --cpu. Pick one."; exit 1; }
+            [[ -n "$MODEL_MODE" ]] && { err "Cannot combine --gpu, --cpu, and --hosted. Pick one."; exit 1; }
             MODEL_MODE="cpu" ;;
+        --hosted)
+            [[ -n "$MODEL_MODE" ]] && { err "Cannot combine --gpu, --cpu, and --hosted. Pick one."; exit 1; }
+            MODEL_MODE="hosted" ;;
         --ollama-gpu)
             [[ -n "$OLLAMA_MODE" ]] && { err "Cannot combine --ollama-gpu and --ollama-cpu. Pick one."; exit 1; }
             OLLAMA_MODE="ollama-gpu" ;;
@@ -224,20 +229,21 @@ for i in "${!ARGS[@]}"; do
             SKIP_NEXT=true ;;
         *)
             err "Unknown argument: $arg"
-            err "Usage: $0 <--gpu|--cpu> [--ollama-gpu|--ollama-cpu] [--llm-model MODEL] [--garage] [--caddy] [--domain DOMAIN] [--password PASS] [--build]"
+            err "Usage: $0 <--gpu|--cpu|--hosted> [--ollama-gpu|--ollama-cpu] [--llm-model MODEL] [--garage] [--caddy] [--domain DOMAIN] [--password PASS] [--build]"
             exit 1
             ;;
     esac
 done
 
 if [[ -z "$MODEL_MODE" ]]; then
-    err "No model mode specified. You must choose --gpu or --cpu."
+    err "No model mode specified. You must choose --gpu, --cpu, or --hosted."
     err ""
-    err "Usage: $0 <--gpu|--cpu> [--ollama-gpu|--ollama-cpu] [--llm-model MODEL] [--garage] [--caddy] [--domain DOMAIN] [--password PASS] [--build]"
+    err "Usage: $0 <--gpu|--cpu|--hosted> [--ollama-gpu|--ollama-cpu] [--llm-model MODEL] [--garage] [--caddy] [--domain DOMAIN] [--password PASS] [--build]"
     err ""
-    err "Specialized models (required):"
+    err "ML processing modes (required):"
-    err "  --gpu              NVIDIA GPU for transcription/diarization/translation"
+    err "  --gpu              NVIDIA GPU container for transcription/diarization/translation"
-    err "  --cpu              CPU-only (slower but works without GPU)"
+    err "  --cpu              In-process CPU processing (no ML container, slower)"
+    err "  --hosted           Remote GPU service URL (no ML container)"
     err ""
     err "Local LLM (optional):"
     err "  --ollama-gpu       Local Ollama with GPU (for summarization/topics)"
@@ -255,7 +261,9 @@ if [[ -z "$MODEL_MODE" ]]; then
 fi
 
 # Build profiles list — one profile per feature
-COMPOSE_PROFILES=("$MODEL_MODE")
+# Only --gpu needs a compose profile; --cpu and --hosted use in-process/remote backends
+COMPOSE_PROFILES=()
+[[ "$MODEL_MODE" == "gpu" ]] && COMPOSE_PROFILES+=("gpu")
 [[ -n "$OLLAMA_MODE" ]] && COMPOSE_PROFILES+=("$OLLAMA_MODE")
 [[ "$USE_GARAGE" == "true" ]] && COMPOSE_PROFILES+=("garage")
 [[ "$USE_CADDY" == "true" ]] && COMPOSE_PROFILES+=("caddy")
@@ -422,43 +430,102 @@ step_server_env() {
         env_set "$SERVER_ENV" "WEBRTC_HOST" "$PRIMARY_IP"
     fi
 
-    # Specialized models (always via gpu/cpu container aliased as "transcription")
+    # Specialized models — backend configuration per mode
-    env_set "$SERVER_ENV" "TRANSCRIPT_BACKEND" "modal"
-    env_set "$SERVER_ENV" "TRANSCRIPT_URL" "http://transcription:8000"
-    env_set "$SERVER_ENV" "TRANSCRIPT_MODAL_API_KEY" "selfhosted"
     env_set "$SERVER_ENV" "DIARIZATION_ENABLED" "true"
-    env_set "$SERVER_ENV" "DIARIZATION_BACKEND" "modal"
+    case "$MODEL_MODE" in
-    env_set "$SERVER_ENV" "DIARIZATION_URL" "http://transcription:8000"
+        gpu)
-    env_set "$SERVER_ENV" "TRANSLATION_BACKEND" "modal"
+            # GPU container aliased as "transcription" on docker network
-    env_set "$SERVER_ENV" "TRANSLATE_URL" "http://transcription:8000"
+            env_set "$SERVER_ENV" "TRANSCRIPT_BACKEND" "modal"
-    env_set "$SERVER_ENV" "PADDING_BACKEND" "modal"
+            env_set "$SERVER_ENV" "TRANSCRIPT_URL" "http://transcription:8000"
-    env_set "$SERVER_ENV" "PADDING_URL" "http://transcription:8000"
+            env_set "$SERVER_ENV" "TRANSCRIPT_MODAL_API_KEY" "selfhosted"
+            env_set "$SERVER_ENV" "DIARIZATION_BACKEND" "modal"
+            env_set "$SERVER_ENV" "DIARIZATION_URL" "http://transcription:8000"
+            env_set "$SERVER_ENV" "TRANSLATION_BACKEND" "modal"
+            env_set "$SERVER_ENV" "TRANSLATE_URL" "http://transcription:8000"
+            env_set "$SERVER_ENV" "PADDING_BACKEND" "modal"
+            env_set "$SERVER_ENV" "PADDING_URL" "http://transcription:8000"
+            ok "ML backends: GPU container (modal)"
+            ;;
+        cpu)
+            # In-process backends — no ML service container needed
+            env_set "$SERVER_ENV" "TRANSCRIPT_BACKEND" "whisper"
+            env_set "$SERVER_ENV" "DIARIZATION_BACKEND" "pyannote"
+            env_set "$SERVER_ENV" "TRANSLATION_BACKEND" "marian"
+            env_set "$SERVER_ENV" "PADDING_BACKEND" "pyav"
+            ok "ML backends: in-process CPU (whisper/pyannote/marian/pyav)"
+            ;;
+        hosted)
+            # Remote GPU service — user provides URL
+            local gpu_url=""
+            if env_has_key "$SERVER_ENV" "TRANSCRIPT_URL"; then
+                gpu_url=$(env_get "$SERVER_ENV" "TRANSCRIPT_URL")
+            fi
+            if [[ -z "$gpu_url" ]] && [[ -t 0 ]]; then
+                echo ""
+                info "Enter the URL of your remote GPU service (e.g. https://gpu.example.com)"
+                read -rp "  GPU service URL: " gpu_url
+            fi
+            if [[ -z "$gpu_url" ]]; then
+                err "GPU service URL required for --hosted mode."
+                err "Set TRANSCRIPT_URL in server/.env or provide it interactively."
+                exit 1
+            fi
+            env_set "$SERVER_ENV" "TRANSCRIPT_BACKEND" "modal"
+            env_set "$SERVER_ENV" "TRANSCRIPT_URL" "$gpu_url"
+            env_set "$SERVER_ENV" "DIARIZATION_BACKEND" "modal"
+            env_set "$SERVER_ENV" "DIARIZATION_URL" "$gpu_url"
+            env_set "$SERVER_ENV" "TRANSLATION_BACKEND" "modal"
+            env_set "$SERVER_ENV" "TRANSLATE_URL" "$gpu_url"
+            env_set "$SERVER_ENV" "PADDING_BACKEND" "modal"
+            env_set "$SERVER_ENV" "PADDING_URL" "$gpu_url"
+            # API key for remote service
+            local gpu_api_key=""
+            if env_has_key "$SERVER_ENV" "TRANSCRIPT_MODAL_API_KEY"; then
+                gpu_api_key=$(env_get "$SERVER_ENV" "TRANSCRIPT_MODAL_API_KEY")
+            fi
+            if [[ -z "$gpu_api_key" ]] && [[ -t 0 ]]; then
+                read -rp "  GPU service API key (or Enter to skip): " gpu_api_key
+            fi
+            if [[ -n "$gpu_api_key" ]]; then
+                env_set "$SERVER_ENV" "TRANSCRIPT_MODAL_API_KEY" "$gpu_api_key"
+            fi
+            ok "ML backends: remote hosted ($gpu_url)"
+            ;;
+    esac
 
     # HuggingFace token for gated models (pyannote diarization)
-    # Written to root .env so docker compose picks it up for gpu/cpu containers
+    # --gpu: written to root .env (docker compose passes to GPU container)
-    local root_env="$ROOT_DIR/.env"
+    # --cpu: written to both root .env and server/.env (in-process pyannote needs it)
-    local current_hf_token="${HF_TOKEN:-}"
+    # --hosted: not needed (remote service handles its own auth)
-    if [[ -f "$root_env" ]] && env_has_key "$root_env" "HF_TOKEN"; then
+    if [[ "$MODEL_MODE" != "hosted" ]]; then
-        current_hf_token=$(env_get "$root_env" "HF_TOKEN")
+        local root_env="$ROOT_DIR/.env"
-    fi
+        local current_hf_token="${HF_TOKEN:-}"
-    if [[ -z "$current_hf_token" ]]; then
+        if [[ -f "$root_env" ]] && env_has_key "$root_env" "HF_TOKEN"; then
-        echo ""
+            current_hf_token=$(env_get "$root_env" "HF_TOKEN")
-        warn "HF_TOKEN not set. Diarization will use a public model fallback."
+        fi
-        warn "For best results, get a token at https://huggingface.co/settings/tokens"
+        if [[ -z "$current_hf_token" ]]; then
-        warn "and accept pyannote licenses at https://huggingface.co/pyannote/speaker-diarization-3.1"
+            echo ""
-        if [[ -t 0 ]]; then
+            warn "HF_TOKEN not set. Diarization will use a public model fallback."
-            read -rp "  HuggingFace token (or press Enter to skip): " current_hf_token
+            warn "For best results, get a token at https://huggingface.co/settings/tokens"
+            warn "and accept pyannote licenses at https://huggingface.co/pyannote/speaker-diarization-3.1"
+            if [[ -t 0 ]]; then
+                read -rp "  HuggingFace token (or press Enter to skip): " current_hf_token
+            fi
+        fi
+        if [[ -n "$current_hf_token" ]]; then
+            touch "$root_env"
+            env_set "$root_env" "HF_TOKEN" "$current_hf_token"
+            export HF_TOKEN="$current_hf_token"
+            # In CPU mode, server process needs HF_TOKEN directly
+            if [[ "$MODEL_MODE" == "cpu" ]]; then
+                env_set "$SERVER_ENV" "HF_TOKEN" "$current_hf_token"
+            fi
+            ok "HF_TOKEN configured"
+        else
+            touch "$root_env"
+            env_set "$root_env" "HF_TOKEN" ""
+            ok "HF_TOKEN skipped (using public model fallback)"
         fi
-    fi
-    if [[ -n "$current_hf_token" ]]; then
-        touch "$root_env"
-        env_set "$root_env" "HF_TOKEN" "$current_hf_token"
-        export HF_TOKEN="$current_hf_token"
-        ok "HF_TOKEN configured"
-    else
-        touch "$root_env"
-        env_set "$root_env" "HF_TOKEN" ""
-        ok "HF_TOKEN skipped (using public model fallback)"
     fi
 
     # LLM configuration
@@ -799,11 +866,12 @@ CADDYEOF
 step_services() {
     info "Step 6: Starting Docker services"
 
-    # Build GPU/CPU image from source (always needed — no prebuilt image)
+    # Build GPU image from source (only for --gpu mode)
-    local build_svc="$MODEL_MODE"
+    if [[ "$MODEL_MODE" == "gpu" ]]; then
-    info "Building $build_svc image (first build downloads ML models, may take a while)..."
+        info "Building gpu image (first build downloads ML models, may take a while)..."
-    compose_cmd build "$build_svc"
+        compose_cmd build gpu
-    ok "$build_svc image built"
+        ok "gpu image built"
+    fi
 
     # Build or pull backend and frontend images
     if [[ "$BUILD_IMAGES" == "true" ]]; then
@@ -871,25 +939,29 @@ step_services() {
 step_health() {
     info "Step 7: Health checks"
 
-    # Specialized model service (gpu or cpu)
+    # Specialized model service (only for --gpu mode)
-    local model_svc="$MODEL_MODE"
+    if [[ "$MODEL_MODE" == "gpu" ]]; then
-
+        info "Waiting for gpu service (first start downloads ~1GB of models)..."
-    info "Waiting for $model_svc service (first start downloads ~1GB of models)..."
+        local model_ok=false
-    local model_ok=false
+        for i in $(seq 1 120); do
-    for i in $(seq 1 120); do
+            if curl -sf http://localhost:8000/docs > /dev/null 2>&1; then
-        if curl -sf http://localhost:8000/docs > /dev/null 2>&1; then
+                model_ok=true
-            model_ok=true
+                break
-            break
+            fi
+            echo -ne "\r  Waiting for gpu service... ($i/120)"
+            sleep 5
+        done
+        echo ""
+        if [[ "$model_ok" == "true" ]]; then
+            ok "gpu service healthy (transcription + diarization)"
+        else
+            warn "gpu service not ready yet — it will keep loading in the background"
+            warn "Check with: docker compose -f docker-compose.selfhosted.yml logs gpu"
         fi
-        echo -ne "\r  Waiting for $model_svc service... ($i/120)"
+    elif [[ "$MODEL_MODE" == "cpu" ]]; then
-        sleep 5
+        ok "CPU mode — ML processing runs in-process on server/worker (no separate service)"
-    done
+    elif [[ "$MODEL_MODE" == "hosted" ]]; then
-    echo ""
+        ok "Hosted mode — ML processing via remote GPU service (no local health check)"
-    if [[ "$model_ok" == "true" ]]; then
-        ok "$model_svc service healthy (transcription + diarization)"
-    else
-        warn "$model_svc service not ready yet — it will keep loading in the background"
-        warn "Check with: docker compose -f docker-compose.selfhosted.yml logs $model_svc"
     fi
 
     # Ollama (if applicable)

server/.env.example

@@ -89,11 +89,11 @@ LLM_CONTEXT_WINDOW=16000
 ## =======================================================
 ## Audio Padding
 ##
-## backends: local (in-process PyAV), modal (HTTP API client)
+## backends: pyav (in-process PyAV), modal (HTTP API client)
-## Default is "local" — no external service needed.
+## Default is "pyav" — no external service needed.
 ## Set to "modal" when using Modal.com or self-hosted gpu/self_hosted/ container.
 ## =======================================================
-#PADDING_BACKEND=local
+#PADDING_BACKEND=pyav
 #PADDING_BACKEND=modal
 #PADDING_URL=https://xxxxx--reflector-padding-web.modal.run
 #PADDING_MODAL_API_KEY=xxxxx
@@ -101,8 +101,8 @@ LLM_CONTEXT_WINDOW=16000
 ## =======================================================
 ## Diarization
 ##
-## Only available on modal
+## backends: modal (HTTP API), pyannote (in-process pyannote.audio)
-## To allow diarization, you need to expose expose the files to be dowloded by the pipeline
+## To allow diarization, you need to expose expose the files to be downloaded by the pipeline
 ## =======================================================
 DIARIZATION_ENABLED=false
 DIARIZATION_BACKEND=modal

server/.env.selfhosted.example

@@ -32,26 +32,46 @@ AUTH_BACKEND=none
 
 # =======================================================
 # Specialized Models (Transcription, Diarization, Translation)
-# These run in the gpu/cpu container — NOT an LLM.
+# These do NOT use an LLM. Configured per mode by the setup script:
-# The "modal" backend means "HTTP API client" — it talks to
+#
-# the self-hosted container, not Modal.com cloud.
+# --gpu mode:    modal backends → GPU container (http://transcription:8000)
+# --cpu mode:    whisper/pyannote/marian/pyav → in-process ML on server/worker
+# --hosted mode: modal backends → user-provided remote GPU service URL
 # =======================================================
+
+# --- --gpu mode (default) ---
 TRANSCRIPT_BACKEND=modal
 TRANSCRIPT_URL=http://transcription:8000
 TRANSCRIPT_MODAL_API_KEY=selfhosted
-
 DIARIZATION_ENABLED=true
 DIARIZATION_BACKEND=modal
 DIARIZATION_URL=http://transcription:8000
-
 TRANSLATION_BACKEND=modal
 TRANSLATE_URL=http://transcription:8000
-
 PADDING_BACKEND=modal
 PADDING_URL=http://transcription:8000
 
-# HuggingFace token — optional, for gated models (e.g. pyannote).
+# --- --cpu mode (set by setup script) ---
-# Falls back to public S3 model bundle if not set.
+# TRANSCRIPT_BACKEND=whisper
+# DIARIZATION_BACKEND=pyannote
+# TRANSLATION_BACKEND=marian
+# PADDING_BACKEND=pyav
+
+# --- --hosted mode (set by setup script) ---
+# TRANSCRIPT_BACKEND=modal
+# TRANSCRIPT_URL=https://your-gpu-service.example.com
+# DIARIZATION_BACKEND=modal
+# DIARIZATION_URL=https://your-gpu-service.example.com
+# ... (all URLs point to one remote service)
+
+# Whisper model sizes for local transcription (--cpu mode)
+# Options: "tiny", "base", "small", "medium", "large-v2"
+# WHISPER_CHUNK_MODEL=tiny
+# WHISPER_FILE_MODEL=tiny
+
+# HuggingFace token — for gated models (e.g. pyannote diarization).
+# Required for --gpu and --cpu modes; falls back to public S3 bundle if not set.
+# Not needed for --hosted mode (remote service handles its own auth).
 # HF_TOKEN=hf_xxxxx
 
 # =======================================================

server/Dockerfile

@@ -6,7 +6,7 @@ ENV PYTHONUNBUFFERED=1 \
 
 # builder install base dependencies
 WORKDIR /tmp
-RUN apt-get update && apt-get install -y curl && apt-get clean
+RUN apt-get update && apt-get install -y curl ffmpeg && apt-get clean
 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"

server/pyproject.toml

@@ -38,7 +38,7 @@ dependencies = [
     "pytest-env>=1.1.5",
     "webvtt-py>=0.5.0",
     "icalendar>=6.0.0",
-    "hatchet-sdk>=0.47.0",
+    "hatchet-sdk==1.22.16",
     "pydantic>=2.12.5",
 ]
 
@@ -71,9 +71,12 @@ local = [
     "faster-whisper>=0.10.0",
 ]
 silero-vad = [
-    "silero-vad>=5.1.2",
+    "silero-vad==5.1.2",
     "torch>=2.8.0",
     "torchaudio>=2.8.0",
+    "pyannote.audio==3.4.0",
+    "pytorch-lightning<2.6",
+    "librosa==0.10.1",
 ]
 
 [tool.uv]

server/reflector/auth/__init__.py

@@ -12,8 +12,10 @@ AccessTokenInfo = auth_module.AccessTokenInfo
 authenticated = auth_module.authenticated
 current_user = auth_module.current_user
 current_user_optional = auth_module.current_user_optional
+current_user_optional_if_public_mode = auth_module.current_user_optional_if_public_mode
 parse_ws_bearer_token = auth_module.parse_ws_bearer_token
 current_user_ws_optional = auth_module.current_user_ws_optional
+verify_raw_token = auth_module.verify_raw_token
 
 # Optional router (e.g. for /auth/login in password backend)
 router = getattr(auth_module, "router", None)

server/reflector/auth/auth_jwt.py

@@ -129,6 +129,17 @@ async def current_user_optional(
     return await _authenticate_user(jwt_token, api_key, jwtauth)
 
 
+async def current_user_optional_if_public_mode(
+    jwt_token: Annotated[Optional[str], Depends(oauth2_scheme)],
+    api_key: Annotated[Optional[str], Depends(api_key_header)],
+    jwtauth: JWTAuth = Depends(),
+) -> Optional[UserInfo]:
+    user = await _authenticate_user(jwt_token, api_key, jwtauth)
+    if user is None and not settings.PUBLIC_MODE:
+        raise HTTPException(status_code=401, detail="Not authenticated")
+    return user
+
+
 def parse_ws_bearer_token(
     websocket: "WebSocket",
 ) -> tuple[Optional[str], Optional[str]]:
@@ -144,3 +155,8 @@ async def current_user_ws_optional(websocket: "WebSocket") -> Optional[UserInfo]
     if not token:
         return None
     return await _authenticate_user(token, None, JWTAuth())
+
+
+async def verify_raw_token(token: str) -> Optional[UserInfo]:
+    """Verify a raw JWT token string (used for query-param auth fallback)."""
+    return await _authenticate_user(token, None, JWTAuth())

server/reflector/auth/auth_none.py

@@ -21,9 +21,19 @@ def current_user_optional():
     return None
 
 
+def current_user_optional_if_public_mode():
+    # auth_none means no authentication at all — always public
+    return None
+
+
 def parse_ws_bearer_token(websocket):
     return None, None
 
 
 async def current_user_ws_optional(websocket):
     return None
+
+
+async def verify_raw_token(token):
+    """Verify a raw JWT token string (used for query-param auth fallback)."""
+    return None

server/reflector/auth/auth_password.py

@@ -150,6 +150,16 @@ async def current_user_optional(
     return await _authenticate_user(jwt_token, api_key)
 
 
+async def current_user_optional_if_public_mode(
+    jwt_token: Annotated[Optional[str], Depends(oauth2_scheme)],
+    api_key: Annotated[Optional[str], Depends(api_key_header)],
+) -> Optional[UserInfo]:
+    user = await _authenticate_user(jwt_token, api_key)
+    if user is None and not settings.PUBLIC_MODE:
+        raise HTTPException(status_code=401, detail="Not authenticated")
+    return user
+
+
 # --- WebSocket auth (same pattern as auth_jwt.py) ---
 def parse_ws_bearer_token(
     websocket: "WebSocket",
@@ -168,6 +178,11 @@ async def current_user_ws_optional(websocket: "WebSocket") -> Optional[UserInfo]
     return await _authenticate_user(token, None)
 
 
+async def verify_raw_token(token: str) -> Optional[UserInfo]:
+    """Verify a raw JWT token string (used for query-param auth fallback)."""
+    return await _authenticate_user(token, None)
+
+
 # --- Login router ---
 router = APIRouter(prefix="/auth", tags=["auth"])
 

server/reflector/db/transcripts.py

@@ -697,6 +697,18 @@ class TranscriptController:
             return False
         return user_id and transcript.user_id == user_id
 
+    @staticmethod
+    def check_can_mutate(transcript: Transcript, user_id: str | None) -> None:
+        """
+        Raises HTTP 403 if the user cannot mutate the transcript.
+
+        Policy:
+        - Anonymous transcripts (user_id is None) are editable by anyone
+        - Owned transcripts can only be mutated by their owner
+        """
+        if transcript.user_id is not None and transcript.user_id != user_id:
+            raise HTTPException(status_code=403, detail="Not authorized")
+
     @asynccontextmanager
     async def transaction(self):
         """

server/reflector/hatchet/constants.py

@@ -39,5 +39,12 @@ TIMEOUT_MEDIUM = (
     300  # Single LLM calls, waveform generation (5m for slow LLM responses)
 )
 TIMEOUT_LONG = 180  # Action items (larger context LLM)
-TIMEOUT_AUDIO = 720  # Audio processing: padding, mixdown
+TIMEOUT_TITLE = 300  # generate_title (single LLM call; doc: reduce from 600s)
-TIMEOUT_HEAVY = 600  # Transcription, fan-out LLM tasks
+TIMEOUT_AUDIO = 720  # Audio processing: padding, mixdown (Hatchet execution_timeout)
+TIMEOUT_AUDIO_HTTP = (
+    660  # httpx timeout for pad_track — below 720 so Hatchet doesn't race
+)
+TIMEOUT_HEAVY = 600  # Transcription, fan-out LLM tasks (Hatchet execution_timeout)
+TIMEOUT_HEAVY_HTTP = (
+    540  # httpx timeout for transcribe_track — below 600 so Hatchet doesn't race
+)

server/reflector/hatchet/error_classification.py

@@ -0,0 +1,74 @@
+"""Classify exceptions as non-retryable for Hatchet workflows.
+
+When a task raises NonRetryableException (or an exception classified as
+non-retryable and re-raised as such), Hatchet stops immediately — no further
+retries. Used by with_error_handling to avoid wasting retries on config errors,
+auth failures, corrupt data, etc.
+"""
+
+# Optional dependencies: only classify if the exception type is available.
+# This avoids hard dependency on openai/av/botocore for code paths that don't use them.
+try:
+    import openai
+except ImportError:
+    openai = None  # type: ignore[assignment]
+
+try:
+    import av
+except ImportError:
+    av = None  # type: ignore[assignment]
+
+try:
+    from botocore.exceptions import ClientError as BotoClientError
+except ImportError:
+    BotoClientError = None  # type: ignore[misc, assignment]
+
+from hatchet_sdk import NonRetryableException
+from httpx import HTTPStatusError
+
+from reflector.llm import LLMParseError
+
+# HTTP status codes that won't change on retry (auth, not found, payment, payload)
+NON_RETRYABLE_HTTP_STATUSES = {401, 402, 403, 404, 413}
+NON_RETRYABLE_S3_CODES = {"AccessDenied", "NoSuchBucket", "NoSuchKey"}
+
+
+def is_non_retryable(e: BaseException) -> bool:
+    """Return True if the exception should stop Hatchet retries immediately.
+
+    Hard failures (config, auth, missing resource, corrupt data) return True.
+    Transient errors (timeouts, 5xx, 429, connection) return False.
+    """
+    if isinstance(e, NonRetryableException):
+        return True
+
+    # Config/input errors
+    if isinstance(e, (ValueError, TypeError)):
+        return True
+
+    # HTTP status codes that won't change on retry
+    if isinstance(e, HTTPStatusError):
+        return e.response.status_code in NON_RETRYABLE_HTTP_STATUSES
+
+    # OpenAI auth errors
+    if openai is not None and isinstance(e, openai.AuthenticationError):
+        return True
+
+    # LLM parse failures (already retried internally)
+    if isinstance(e, LLMParseError):
+        return True
+
+    # S3 permission/existence errors
+    if BotoClientError is not None and isinstance(e, BotoClientError):
+        code = e.response.get("Error", {}).get("Code", "")
+        return code in NON_RETRYABLE_S3_CODES
+
+    # Corrupt audio (PyAV) — AVError in some versions; fallback to InvalidDataError
+    if av is not None:
+        av_error = getattr(av, "AVError", None) or getattr(
+            getattr(av, "error", None), "InvalidDataError", None
+        )
+        if av_error is not None and isinstance(e, av_error):
+            return True
+
+    return False

server/reflector/hatchet/workflows/daily_multitrack_pipeline.py

@@ -27,6 +27,7 @@ from hatchet_sdk import (
     ConcurrencyExpression,
     ConcurrencyLimitStrategy,
     Context,
+    NonRetryableException,
 )
 from hatchet_sdk.labels import DesiredWorkerLabel
 from pydantic import BaseModel
@@ -43,8 +44,10 @@ from reflector.hatchet.constants import (
     TIMEOUT_LONG,
     TIMEOUT_MEDIUM,
     TIMEOUT_SHORT,
+    TIMEOUT_TITLE,
     TaskName,
 )
+from reflector.hatchet.error_classification import is_non_retryable
 from reflector.hatchet.workflows.models import (
     ActionItemsResult,
     ConsentResult,
@@ -216,6 +219,13 @@ def make_audio_progress_logger(
 R = TypeVar("R")
 
 
+def _successful_run_results(
+    results: list[dict[str, Any] | BaseException],
+) -> list[dict[str, Any]]:
+    """Return only successful (non-exception) results from aio_run_many(return_exceptions=True)."""
+    return [r for r in results if not isinstance(r, BaseException)]
+
+
 def with_error_handling(
     step_name: TaskName, set_error_status: bool = True
 ) -> Callable[
@@ -243,8 +253,12 @@ def with_error_handling(
                     error=str(e),
                     exc_info=True,
                 )
-                if set_error_status:
+                if is_non_retryable(e):
-                    await set_workflow_error_status(input.transcript_id)
+                    # Hard fail: stop retries, set error status, fail workflow
+                    if set_error_status:
+                        await set_workflow_error_status(input.transcript_id)
+                    raise NonRetryableException(str(e)) from e
+                # Transient: do not set error status — Hatchet will retry
                 raise
 
         return wrapper  # type: ignore[return-value]
@@ -253,7 +267,10 @@ def with_error_handling(
 
 
 @daily_multitrack_pipeline.task(
-    execution_timeout=timedelta(seconds=TIMEOUT_SHORT), retries=3
+    execution_timeout=timedelta(seconds=TIMEOUT_SHORT),
+    retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=10,
 )
 @with_error_handling(TaskName.GET_RECORDING)
 async def get_recording(input: PipelineInput, ctx: Context) -> RecordingResult:
@@ -309,6 +326,8 @@ async def get_recording(input: PipelineInput, ctx: Context) -> RecordingResult:
     parents=[get_recording],
     execution_timeout=timedelta(seconds=TIMEOUT_SHORT),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=10,
 )
 @with_error_handling(TaskName.GET_PARTICIPANTS)
 async def get_participants(input: PipelineInput, ctx: Context) -> ParticipantsResult:
@@ -412,6 +431,8 @@ async def get_participants(input: PipelineInput, ctx: Context) -> ParticipantsRe
     parents=[get_participants],
     execution_timeout=timedelta(seconds=TIMEOUT_HEAVY),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=30,
 )
 @with_error_handling(TaskName.PROCESS_TRACKS)
 async def process_tracks(input: PipelineInput, ctx: Context) -> ProcessTracksResult:
@@ -435,7 +456,7 @@ async def process_tracks(input: PipelineInput, ctx: Context) -> ProcessTracksRes
         for i, track in enumerate(input.tracks)
     ]
 
-    results = await track_workflow.aio_run_many(bulk_runs)
+    results = await track_workflow.aio_run_many(bulk_runs, return_exceptions=True)
 
     target_language = participants_result.target_language
 
@@ -443,7 +464,18 @@ async def process_tracks(input: PipelineInput, ctx: Context) -> ProcessTracksRes
     padded_tracks = []
     created_padded_files = set()
 
-    for result in results:
+    for i, result in enumerate(results):
+        if isinstance(result, BaseException):
+            logger.error(
+                "[Hatchet] process_tracks: track workflow failed, failing step",
+                transcript_id=input.transcript_id,
+                track_index=i,
+                error=str(result),
+            )
+            ctx.log(f"process_tracks: track {i} failed ({result}), failing step")
+            raise ValueError(
+                f"Track {i} workflow failed after retries: {result!s}"
+            ) from result
         transcribe_result = TranscribeTrackResult(**result[TaskName.TRANSCRIBE_TRACK])
         track_words.append(transcribe_result.words)
 
@@ -481,7 +513,9 @@ async def process_tracks(input: PipelineInput, ctx: Context) -> ProcessTracksRes
 @daily_multitrack_pipeline.task(
     parents=[process_tracks],
     execution_timeout=timedelta(seconds=TIMEOUT_AUDIO),
-    retries=3,
+    retries=2,
+    backoff_factor=2.0,
+    backoff_max_seconds=15,
     desired_worker_labels={
         "pool": DesiredWorkerLabel(
             value="cpu-heavy",
@@ -593,6 +627,8 @@ async def mixdown_tracks(input: PipelineInput, ctx: Context) -> MixdownResult:
     parents=[mixdown_tracks],
     execution_timeout=timedelta(seconds=TIMEOUT_MEDIUM),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=10,
 )
 @with_error_handling(TaskName.GENERATE_WAVEFORM)
 async def generate_waveform(input: PipelineInput, ctx: Context) -> WaveformResult:
@@ -661,6 +697,8 @@ async def generate_waveform(input: PipelineInput, ctx: Context) -> WaveformResul
     parents=[process_tracks],
     execution_timeout=timedelta(seconds=TIMEOUT_HEAVY),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=30,
 )
 @with_error_handling(TaskName.DETECT_TOPICS)
 async def detect_topics(input: PipelineInput, ctx: Context) -> TopicsResult:
@@ -722,11 +760,22 @@ async def detect_topics(input: PipelineInput, ctx: Context) -> TopicsResult:
         for chunk in chunks
     ]
 
-    results = await topic_chunk_workflow.aio_run_many(bulk_runs)
+    results = await topic_chunk_workflow.aio_run_many(bulk_runs, return_exceptions=True)
 
-    topic_chunks = [
+    topic_chunks: list[TopicChunkResult] = []
-        TopicChunkResult(**result[TaskName.DETECT_CHUNK_TOPIC]) for result in results
+    for i, result in enumerate(results):
-    ]
+        if isinstance(result, BaseException):
+            logger.error(
+                "[Hatchet] detect_topics: chunk workflow failed, failing step",
+                transcript_id=input.transcript_id,
+                chunk_index=i,
+                error=str(result),
+            )
+            ctx.log(f"detect_topics: chunk {i} failed ({result}), failing step")
+            raise ValueError(
+                f"Topic chunk {i} workflow failed after retries: {result!s}"
+            ) from result
+        topic_chunks.append(TopicChunkResult(**result[TaskName.DETECT_CHUNK_TOPIC]))
 
     async with fresh_db_connection():
         transcript = await transcripts_controller.get_by_id(input.transcript_id)
@@ -764,8 +813,10 @@ async def detect_topics(input: PipelineInput, ctx: Context) -> TopicsResult:
 
 @daily_multitrack_pipeline.task(
     parents=[detect_topics],
-    execution_timeout=timedelta(seconds=TIMEOUT_HEAVY),
+    execution_timeout=timedelta(seconds=TIMEOUT_TITLE),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=15,
 )
 @with_error_handling(TaskName.GENERATE_TITLE)
 async def generate_title(input: PipelineInput, ctx: Context) -> TitleResult:
@@ -830,7 +881,9 @@ async def generate_title(input: PipelineInput, ctx: Context) -> TitleResult:
 @daily_multitrack_pipeline.task(
     parents=[detect_topics],
     execution_timeout=timedelta(seconds=TIMEOUT_MEDIUM),
-    retries=3,
+    retries=5,
+    backoff_factor=2.0,
+    backoff_max_seconds=30,
 )
 @with_error_handling(TaskName.EXTRACT_SUBJECTS)
 async def extract_subjects(input: PipelineInput, ctx: Context) -> SubjectsResult:
@@ -909,6 +962,8 @@ async def extract_subjects(input: PipelineInput, ctx: Context) -> SubjectsResult
     parents=[extract_subjects],
     execution_timeout=timedelta(seconds=TIMEOUT_HEAVY),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=30,
 )
 @with_error_handling(TaskName.PROCESS_SUBJECTS)
 async def process_subjects(input: PipelineInput, ctx: Context) -> ProcessSubjectsResult:
@@ -935,12 +990,24 @@ async def process_subjects(input: PipelineInput, ctx: Context) -> ProcessSubject
         for i, subject in enumerate(subjects)
     ]
 
-    results = await subject_workflow.aio_run_many(bulk_runs)
+    results = await subject_workflow.aio_run_many(bulk_runs, return_exceptions=True)
 
-    subject_summaries = [
+    subject_summaries: list[SubjectSummaryResult] = []
-        SubjectSummaryResult(**result[TaskName.GENERATE_DETAILED_SUMMARY])
+    for i, result in enumerate(results):
-        for result in results
+        if isinstance(result, BaseException):
-    ]
+            logger.error(
+                "[Hatchet] process_subjects: subject workflow failed, failing step",
+                transcript_id=input.transcript_id,
+                subject_index=i,
+                error=str(result),
+            )
+            ctx.log(f"process_subjects: subject {i} failed ({result}), failing step")
+            raise ValueError(
+                f"Subject {i} workflow failed after retries: {result!s}"
+            ) from result
+        subject_summaries.append(
+            SubjectSummaryResult(**result[TaskName.GENERATE_DETAILED_SUMMARY])
+        )
 
     ctx.log(f"process_subjects complete: {len(subject_summaries)} summaries")
 
@@ -951,6 +1018,8 @@ async def process_subjects(input: PipelineInput, ctx: Context) -> ProcessSubject
     parents=[process_subjects],
     execution_timeout=timedelta(seconds=TIMEOUT_MEDIUM),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=15,
 )
 @with_error_handling(TaskName.GENERATE_RECAP)
 async def generate_recap(input: PipelineInput, ctx: Context) -> RecapResult:
@@ -1040,6 +1109,8 @@ async def generate_recap(input: PipelineInput, ctx: Context) -> RecapResult:
     parents=[extract_subjects],
     execution_timeout=timedelta(seconds=TIMEOUT_LONG),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=15,
 )
 @with_error_handling(TaskName.IDENTIFY_ACTION_ITEMS)
 async def identify_action_items(
@@ -1108,6 +1179,8 @@ async def identify_action_items(
     parents=[process_tracks, generate_title, generate_recap, identify_action_items],
     execution_timeout=timedelta(seconds=TIMEOUT_SHORT),
     retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=5,
 )
 @with_error_handling(TaskName.FINALIZE)
 async def finalize(input: PipelineInput, ctx: Context) -> FinalizeResult:
@@ -1177,7 +1250,11 @@ async def finalize(input: PipelineInput, ctx: Context) -> FinalizeResult:
 
 
 @daily_multitrack_pipeline.task(
-    parents=[finalize], execution_timeout=timedelta(seconds=TIMEOUT_SHORT), retries=3
+    parents=[finalize],
+    execution_timeout=timedelta(seconds=TIMEOUT_SHORT),
+    retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=10,
 )
 @with_error_handling(TaskName.CLEANUP_CONSENT, set_error_status=False)
 async def cleanup_consent(input: PipelineInput, ctx: Context) -> ConsentResult:
@@ -1283,6 +1360,8 @@ async def cleanup_consent(input: PipelineInput, ctx: Context) -> ConsentResult:
     parents=[cleanup_consent],
     execution_timeout=timedelta(seconds=TIMEOUT_SHORT),
     retries=5,
+    backoff_factor=2.0,
+    backoff_max_seconds=15,
 )
 @with_error_handling(TaskName.POST_ZULIP, set_error_status=False)
 async def post_zulip(input: PipelineInput, ctx: Context) -> ZulipResult:
@@ -1310,6 +1389,8 @@ async def post_zulip(input: PipelineInput, ctx: Context) -> ZulipResult:
     parents=[cleanup_consent],
     execution_timeout=timedelta(seconds=TIMEOUT_MEDIUM),
     retries=5,
+    backoff_factor=2.0,
+    backoff_max_seconds=15,
 )
 @with_error_handling(TaskName.SEND_WEBHOOK, set_error_status=False)
 async def send_webhook(input: PipelineInput, ctx: Context) -> WebhookResult:
@@ -1378,3 +1459,32 @@ async def send_webhook(input: PipelineInput, ctx: Context) -> WebhookResult:
         except Exception as e:
             ctx.log(f"send_webhook unexpected error, continuing anyway: {e}")
             return WebhookResult(webhook_sent=False)
+
+
+async def on_workflow_failure(input: PipelineInput, ctx: Context) -> None:
+    """Run when the workflow is truly dead (all retries exhausted).
+
+    Sets transcript status to 'error' only if it is not already 'ended'.
+    Post-finalize tasks (cleanup_consent, post_zulip, send_webhook) use
+    set_error_status=False; if one of them fails, we must not overwrite
+    the 'ended' status that finalize already set.
+    """
+    async with fresh_db_connection():
+        from reflector.db.transcripts import transcripts_controller  # noqa: PLC0415
+
+        transcript = await transcripts_controller.get_by_id(input.transcript_id)
+        if transcript and transcript.status == "ended":
+            logger.info(
+                "[Hatchet] on_workflow_failure: transcript already ended, skipping error status (failure was post-finalize)",
+                transcript_id=input.transcript_id,
+            )
+            ctx.log(
+                "on_workflow_failure: transcript already ended, skipping error status"
+            )
+            return
+    await set_workflow_error_status(input.transcript_id)
+
+
+@daily_multitrack_pipeline.on_failure_task()
+async def _register_on_workflow_failure(input: PipelineInput, ctx: Context) -> None:
+    await on_workflow_failure(input, ctx)

server/reflector/hatchet/workflows/padding_workflow.py

@@ -34,7 +34,12 @@ padding_workflow = hatchet.workflow(
 )
 
 
-@padding_workflow.task(execution_timeout=timedelta(seconds=TIMEOUT_AUDIO), retries=3)
+@padding_workflow.task(
+    execution_timeout=timedelta(seconds=TIMEOUT_AUDIO),
+    retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=30,
+)
 async def pad_track(input: PaddingInput, ctx: Context) -> PadTrackResult:
     """Pad audio track with silence based on WebM container start_time."""
     ctx.log(f"pad_track: track {input.track_index}, s3_key={input.s3_key}")

server/reflector/hatchet/workflows/subject_processing.py

@@ -13,7 +13,7 @@ from hatchet_sdk.rate_limit import RateLimit
 from pydantic import BaseModel
 
 from reflector.hatchet.client import HatchetClientManager
-from reflector.hatchet.constants import LLM_RATE_LIMIT_KEY, TIMEOUT_MEDIUM
+from reflector.hatchet.constants import LLM_RATE_LIMIT_KEY, TIMEOUT_HEAVY
 from reflector.hatchet.workflows.models import SubjectSummaryResult
 from reflector.logger import logger
 from reflector.processors.summary.prompts import (
@@ -41,8 +41,10 @@ subject_workflow = hatchet.workflow(
 
 
 @subject_workflow.task(
-    execution_timeout=timedelta(seconds=TIMEOUT_MEDIUM),
+    execution_timeout=timedelta(seconds=TIMEOUT_HEAVY),
-    retries=3,
+    retries=5,
+    backoff_factor=2.0,
+    backoff_max_seconds=60,
     rate_limits=[RateLimit(static_key=LLM_RATE_LIMIT_KEY, units=2)],
 )
 async def generate_detailed_summary(

server/reflector/hatchet/workflows/topic_chunk_processing.py

@@ -50,7 +50,9 @@ topic_chunk_workflow = hatchet.workflow(
 
 @topic_chunk_workflow.task(
     execution_timeout=timedelta(seconds=TIMEOUT_MEDIUM),
-    retries=3,
+    retries=5,
+    backoff_factor=2.0,
+    backoff_max_seconds=60,
     rate_limits=[RateLimit(static_key=LLM_RATE_LIMIT_KEY, units=1)],
 )
 async def detect_chunk_topic(input: TopicChunkInput, ctx: Context) -> TopicChunkResult:

server/reflector/hatchet/workflows/track_processing.py

@@ -44,7 +44,12 @@ hatchet = HatchetClientManager.get_client()
 track_workflow = hatchet.workflow(name="TrackProcessing", input_validator=TrackInput)
 
 
-@track_workflow.task(execution_timeout=timedelta(seconds=TIMEOUT_AUDIO), retries=3)
+@track_workflow.task(
+    execution_timeout=timedelta(seconds=TIMEOUT_AUDIO),
+    retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=30,
+)
 async def pad_track(input: TrackInput, ctx: Context) -> PadTrackResult:
     """Pad single audio track with silence for alignment.
 
@@ -137,7 +142,11 @@ async def pad_track(input: TrackInput, ctx: Context) -> PadTrackResult:
 
 
 @track_workflow.task(
-    parents=[pad_track], execution_timeout=timedelta(seconds=TIMEOUT_HEAVY), retries=3
+    parents=[pad_track],
+    execution_timeout=timedelta(seconds=TIMEOUT_HEAVY),
+    retries=3,
+    backoff_factor=2.0,
+    backoff_max_seconds=30,
 )
 async def transcribe_track(input: TrackInput, ctx: Context) -> TranscribeTrackResult:
     """Transcribe audio track using GPU (Modal.com) or local Whisper."""

server/reflector/llm.py

@@ -65,10 +65,25 @@ class LLM:
     async def get_response(
         self, prompt: str, texts: list[str], tone_name: str | None = None
     ) -> str:
-        """Get a text response using TreeSummarize for non-function-calling models"""
+        """Get a text response using TreeSummarize for non-function-calling models.
-        summarizer = TreeSummarize(verbose=False)
+
-        response = await summarizer.aget_response(prompt, texts, tone_name=tone_name)
+        Uses the same retry() wrapper as get_structured_response for transient
-        return str(response).strip()
+        network errors (connection, timeout, OSError) with exponential backoff.
+        """
+
+        async def _call():
+            summarizer = TreeSummarize(verbose=False)
+            response = await summarizer.aget_response(
+                prompt, texts, tone_name=tone_name
+            )
+            return str(response).strip()
+
+        return await retry(_call)(
+            retry_attempts=3,
+            retry_backoff_interval=1.0,
+            retry_backoff_max=30.0,
+            retry_ignore_exc_types=(ConnectionError, TimeoutError, OSError),
+        )
 
     async def get_structured_response(
         self,

server/reflector/processors/__init__.py

@@ -4,6 +4,8 @@ from .audio_diarization_auto import AudioDiarizationAutoProcessor  # noqa: F401
 from .audio_downscale import AudioDownscaleProcessor  # noqa: F401
 from .audio_file_writer import AudioFileWriterProcessor  # noqa: F401
 from .audio_merge import AudioMergeProcessor  # noqa: F401
+from .audio_padding import AudioPaddingProcessor  # noqa: F401
+from .audio_padding_auto import AudioPaddingAutoProcessor  # noqa: F401
 from .audio_transcript import AudioTranscriptProcessor  # noqa: F401
 from .audio_transcript_auto import AudioTranscriptAutoProcessor  # noqa: F401
 from .base import (  # noqa: F401

server/reflector/processors/_audio_download.py

@@ -0,0 +1,86 @@
+"""
+Shared audio download utility for local processors.
+
+Downloads audio from a URL to a temporary file for in-process ML inference.
+"""
+
+import asyncio
+import os
+import tempfile
+from pathlib import Path
+
+import requests
+
+from reflector.logger import logger
+
+S3_TIMEOUT = 60
+
+
+async def download_audio_to_temp(url: str) -> Path:
+    """Download audio from URL to a temporary file.
+
+    The caller is responsible for deleting the temp file after use.
+
+    Args:
+        url: Presigned URL or public URL to download audio from.
+
+    Returns:
+        Path to the downloaded temporary file.
+    """
+    loop = asyncio.get_event_loop()
+    return await loop.run_in_executor(None, _download_blocking, url)
+
+
+def _download_blocking(url: str) -> Path:
+    """Blocking download implementation."""
+    log = logger.bind(url=url[:80])
+    log.info("Downloading audio to temp file")
+
+    response = requests.get(url, stream=True, timeout=S3_TIMEOUT)
+    response.raise_for_status()
+
+    # Determine extension from content-type or URL
+    ext = _detect_extension(url, response.headers.get("content-type", ""))
+
+    fd, tmp_path = tempfile.mkstemp(suffix=ext)
+    try:
+        total_bytes = 0
+        with os.fdopen(fd, "wb") as f:
+            for chunk in response.iter_content(chunk_size=8192):
+                if chunk:
+                    f.write(chunk)
+                    total_bytes += len(chunk)
+        log.info("Audio downloaded", bytes=total_bytes, path=tmp_path)
+        return Path(tmp_path)
+    except Exception:
+        # Clean up on failure
+        try:
+            os.unlink(tmp_path)
+        except OSError:
+            pass
+        raise
+
+
+def _detect_extension(url: str, content_type: str) -> str:
+    """Detect audio file extension from URL or content-type."""
+    # Try URL path first
+    path = url.split("?")[0]  # Strip query params
+    for ext in (".wav", ".mp3", ".mp4", ".m4a", ".webm", ".ogg", ".flac"):
+        if path.lower().endswith(ext):
+            return ext
+
+    # Try content-type
+    ct_map = {
+        "audio/wav": ".wav",
+        "audio/x-wav": ".wav",
+        "audio/mpeg": ".mp3",
+        "audio/mp4": ".m4a",
+        "audio/webm": ".webm",
+        "audio/ogg": ".ogg",
+        "audio/flac": ".flac",
+    }
+    for ct, ext in ct_map.items():
+        if ct in content_type.lower():
+            return ext
+
+    return ".audio"

server/reflector/processors/_marian_translator_service.py

@@ -0,0 +1,76 @@
+"""
+MarianMT translation service.
+
+Singleton service that loads HuggingFace MarianMT translation models
+and reuses them across all MarianMT translator processor instances.
+
+Ported from gpu/self_hosted/app/services/translator.py for in-process use.
+"""
+
+import logging
+import threading
+
+from transformers import MarianMTModel, MarianTokenizer, pipeline
+
+logger = logging.getLogger(__name__)
+
+
+class MarianTranslatorService:
+    """MarianMT text translation service for in-process use."""
+
+    def __init__(self):
+        self._pipeline = None
+        self._current_pair = None
+        self._lock = threading.Lock()
+
+    def load(self, source_language: str = "en", target_language: str = "fr"):
+        """Load the translation model for a specific language pair."""
+        model_name = self._resolve_model_name(source_language, target_language)
+        logger.info(
+            "Loading MarianMT model: %s (%s -> %s)",
+            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)
+        self._current_pair = (source_language.lower(), target_language.lower())
+
+    def _resolve_model_name(self, src: str, tgt: str) -> str:
+        """Resolve language pair to MarianMT model name."""
+        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:
+        """Translate text between languages.
+
+        Args:
+            text: Text to translate.
+            source_language: Source language code (e.g. "en").
+            target_language: Target language code (e.g. "fr").
+
+        Returns:
+            dict with "text" key containing {source_language: original, target_language: translated}.
+        """
+        pair = (source_language.lower(), target_language.lower())
+        if self._pipeline is None or self._current_pair != pair:
+            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}}
+
+
+# Module-level singleton — shared across all MarianMT translator processors
+translator_service = MarianTranslatorService()

server/reflector/processors/_pyannote_diarization_service.py

@@ -0,0 +1,133 @@
+"""
+Pyannote diarization service using pyannote.audio.
+
+Singleton service that loads the pyannote speaker diarization model once
+and reuses it across all pyannote diarization processor instances.
+
+Ported from gpu/self_hosted/app/services/diarizer.py for in-process use.
+"""
+
+import logging
+import tarfile
+import threading
+from pathlib import Path
+from urllib.request import urlopen
+
+import torch
+import torchaudio
+import yaml
+from pyannote.audio import Pipeline
+
+from reflector.settings import settings
+
+logger = logging.getLogger(__name__)
+
+S3_BUNDLE_URL = "https://reflector-public.s3.us-east-1.amazonaws.com/pyannote-speaker-diarization-3.1.tar.gz"
+BUNDLE_CACHE_DIR = Path.home() / ".cache" / "pyannote-bundle"
+
+
+def _ensure_model(cache_dir: Path) -> str:
+    """Download and extract S3 model bundle if not cached."""
+    model_dir = cache_dir / "pyannote-speaker-diarization-3.1"
+    config_path = model_dir / "config.yaml"
+
+    if config_path.exists():
+        logger.info("Using cached model bundle at %s", model_dir)
+        return str(model_dir)
+
+    cache_dir.mkdir(parents=True, exist_ok=True)
+    tarball_path = cache_dir / "model.tar.gz"
+
+    logger.info("Downloading model bundle from %s", S3_BUNDLE_URL)
+    with urlopen(S3_BUNDLE_URL) as response, open(tarball_path, "wb") as f:
+        while chunk := response.read(8192):
+            f.write(chunk)
+
+    logger.info("Extracting model bundle")
+    with tarfile.open(tarball_path, "r:gz") as tar:
+        tar.extractall(path=cache_dir, filter="data")
+    tarball_path.unlink()
+
+    _patch_config(model_dir, cache_dir)
+    return str(model_dir)
+
+
+def _patch_config(model_dir: Path, cache_dir: Path) -> None:
+    """Rewrite config.yaml to reference local pytorch_model.bin paths."""
+    config_path = model_dir / "config.yaml"
+    with open(config_path) as f:
+        config = yaml.safe_load(f)
+
+    config["pipeline"]["params"]["segmentation"] = str(
+        cache_dir / "pyannote-segmentation-3.0" / "pytorch_model.bin"
+    )
+    config["pipeline"]["params"]["embedding"] = str(
+        cache_dir / "pyannote-wespeaker-voxceleb-resnet34-LM" / "pytorch_model.bin"
+    )
+
+    with open(config_path, "w") as f:
+        yaml.dump(config, f)
+
+    logger.info("Patched config.yaml with local model paths")
+
+
+class PyannoteDiarizationService:
+    """Pyannote speaker diarization service for in-process use."""
+
+    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"
+        hf_token = settings.HF_TOKEN
+
+        if hf_token:
+            logger.info("Loading pyannote model from HuggingFace (HF_TOKEN set)")
+            self._pipeline = Pipeline.from_pretrained(
+                "pyannote/speaker-diarization-3.1",
+                use_auth_token=hf_token,
+            )
+        else:
+            logger.info("HF_TOKEN not set — loading model from S3 bundle")
+            model_path = _ensure_model(BUNDLE_CACHE_DIR)
+            config_path = Path(model_path) / "config.yaml"
+            self._pipeline = Pipeline.from_pretrained(str(config_path))
+
+        self._pipeline.to(torch.device(self._device))
+
+    def diarize_file(self, file_path: str, timestamp: float = 0.0) -> dict:
+        """Run speaker diarization on an audio file.
+
+        Args:
+            file_path: Path to the audio file.
+            timestamp: Offset to add to all segment timestamps.
+
+        Returns:
+            dict with "diarization" key containing list of
+            {"start": float, "end": float, "speaker": int} segments.
+        """
+        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}
+            )
+        segments = []
+        for diarization_segment, _, speaker in diarization.itertracks(yield_label=True):
+            segments.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": segments}
+
+
+# Module-level singleton — shared across all pyannote diarization processors
+diarization_service = PyannoteDiarizationService()

server/reflector/processors/audio_diarization_pyannote.py

@@ -0,0 +1,37 @@
+"""
+Pyannote audio diarization processor using pyannote.audio in-process.
+
+Downloads audio from URL, runs pyannote diarization locally,
+and returns speaker segments. No HTTP backend needed.
+"""
+
+import asyncio
+import os
+
+from reflector.processors._audio_download import download_audio_to_temp
+from reflector.processors._pyannote_diarization_service import diarization_service
+from reflector.processors.audio_diarization import AudioDiarizationProcessor
+from reflector.processors.audio_diarization_auto import AudioDiarizationAutoProcessor
+from reflector.processors.types import AudioDiarizationInput
+
+
+class AudioDiarizationPyannoteProcessor(AudioDiarizationProcessor):
+    INPUT_TYPE = AudioDiarizationInput
+
+    async def _diarize(self, data: AudioDiarizationInput):
+        """Run pyannote diarization on audio from URL."""
+        tmp_path = await download_audio_to_temp(data.audio_url)
+        try:
+            loop = asyncio.get_event_loop()
+            result = await loop.run_in_executor(
+                None, diarization_service.diarize_file, str(tmp_path)
+            )
+            return result["diarization"]
+        finally:
+            try:
+                os.unlink(tmp_path)
+            except OSError:
+                pass
+
+
+AudioDiarizationAutoProcessor.register("pyannote", AudioDiarizationPyannoteProcessor)

server/reflector/processors/audio_padding.py

@@ -0,0 +1,23 @@
+"""
+Base class for audio padding processors.
+"""
+
+from pydantic import BaseModel
+
+
+class PaddingResponse(BaseModel):
+    size: int
+    cancelled: bool = False
+
+
+class AudioPaddingProcessor:
+    """Base class for audio padding processors."""
+
+    async def pad_track(
+        self,
+        track_url: str,
+        output_url: str,
+        start_time_seconds: float,
+        track_index: int,
+    ) -> PaddingResponse:
+        raise NotImplementedError

server/reflector/processors/audio_padding_auto.py

@@ -1,9 +1,10 @@
 import importlib
 
+from reflector.processors.audio_padding import AudioPaddingProcessor
 from reflector.settings import settings
 
 
-class AudioPaddingAutoProcessor:
+class AudioPaddingAutoProcessor(AudioPaddingProcessor):
     _registry = {}
 
     @classmethod

server/reflector/processors/audio_padding_modal.py

@@ -6,19 +6,14 @@ import asyncio
 import os
 
 import httpx
-from pydantic import BaseModel
 
-from reflector.hatchet.constants import TIMEOUT_AUDIO
+from reflector.hatchet.constants import TIMEOUT_AUDIO_HTTP
 from reflector.logger import logger
+from reflector.processors.audio_padding import AudioPaddingProcessor, PaddingResponse
 from reflector.processors.audio_padding_auto import AudioPaddingAutoProcessor
 
 
-class PaddingResponse(BaseModel):
+class AudioPaddingModalProcessor(AudioPaddingProcessor):
-    size: int
-    cancelled: bool = False
-
-
-class AudioPaddingModalProcessor:
     """Audio padding processor using Modal.com CPU backend via HTTP."""
 
     def __init__(
@@ -65,7 +60,7 @@ class AudioPaddingModalProcessor:
             headers["Authorization"] = f"Bearer {self.modal_api_key}"
 
         try:
-            async with httpx.AsyncClient(timeout=TIMEOUT_AUDIO) as client:
+            async with httpx.AsyncClient(timeout=TIMEOUT_AUDIO_HTTP) as client:
                 response = await client.post(
                     url,
                     headers=headers,

server/reflector/processors/audio_padding_pyav.py

@@ -1,5 +1,5 @@
 """
-Local audio padding processor using PyAV.
+PyAV audio padding processor.
 
 Pads audio tracks with silence directly in-process (no HTTP).
 Reuses the shared PyAV utilities from reflector.utils.audio_padding.
@@ -12,15 +12,15 @@ import tempfile
 import av
 
 from reflector.logger import logger
+from reflector.processors.audio_padding import AudioPaddingProcessor, PaddingResponse
 from reflector.processors.audio_padding_auto import AudioPaddingAutoProcessor
-from reflector.processors.audio_padding_modal import PaddingResponse
 from reflector.utils.audio_padding import apply_audio_padding_to_file
 
 S3_TIMEOUT = 60
 
 
-class AudioPaddingLocalProcessor:
+class AudioPaddingPyavProcessor(AudioPaddingProcessor):
-    """Audio padding processor using local PyAV (no HTTP backend)."""
+    """Audio padding processor using PyAV (no HTTP backend)."""
 
     async def pad_track(
         self,
@@ -29,7 +29,7 @@ class AudioPaddingLocalProcessor:
         start_time_seconds: float,
         track_index: int,
     ) -> PaddingResponse:
-        """Pad audio track with silence locally via PyAV.
+        """Pad audio track with silence via PyAV.
 
         Args:
             track_url: Presigned GET URL for source audio track
@@ -130,4 +130,4 @@ class AudioPaddingLocalProcessor:
                 log.warning("Failed to cleanup temp directory", error=str(e))
 
 
-AudioPaddingAutoProcessor.register("local", AudioPaddingLocalProcessor)
+AudioPaddingAutoProcessor.register("pyav", AudioPaddingPyavProcessor)

server/reflector/processors/audio_transcript_whisper.py

@@ -3,13 +3,17 @@ from faster_whisper import WhisperModel
 from reflector.processors.audio_transcript import AudioTranscriptProcessor
 from reflector.processors.audio_transcript_auto import AudioTranscriptAutoProcessor
 from reflector.processors.types import AudioFile, Transcript, Word
+from reflector.settings import settings
 
 
 class AudioTranscriptWhisperProcessor(AudioTranscriptProcessor):
     def __init__(self):
         super().__init__()
         self.model = WhisperModel(
-            "tiny", device="cpu", compute_type="float32", num_workers=12
+            settings.WHISPER_CHUNK_MODEL,
+            device="cpu",
+            compute_type="float32",
+            num_workers=12,
         )
 
     async def _transcript(self, data: AudioFile):

server/reflector/processors/file_diarization_pyannote.py

@@ -0,0 +1,39 @@
+"""
+Pyannote file diarization processor using pyannote.audio in-process.
+
+Downloads audio from URL, runs pyannote diarization locally,
+and returns speaker segments. No HTTP backend needed.
+"""
+
+import asyncio
+import os
+
+from reflector.processors._audio_download import download_audio_to_temp
+from reflector.processors._pyannote_diarization_service import diarization_service
+from reflector.processors.file_diarization import (
+    FileDiarizationInput,
+    FileDiarizationOutput,
+    FileDiarizationProcessor,
+)
+from reflector.processors.file_diarization_auto import FileDiarizationAutoProcessor
+
+
+class FileDiarizationPyannoteProcessor(FileDiarizationProcessor):
+    async def _diarize(self, data: FileDiarizationInput):
+        """Run pyannote diarization on file from URL."""
+        self.logger.info(f"Starting pyannote diarization from {data.audio_url}")
+        tmp_path = await download_audio_to_temp(data.audio_url)
+        try:
+            loop = asyncio.get_event_loop()
+            result = await loop.run_in_executor(
+                None, diarization_service.diarize_file, str(tmp_path)
+            )
+            return FileDiarizationOutput(diarization=result["diarization"])
+        finally:
+            try:
+                os.unlink(tmp_path)
+            except OSError:
+                pass
+
+
+FileDiarizationAutoProcessor.register("pyannote", FileDiarizationPyannoteProcessor)

server/reflector/processors/file_transcript_whisper.py

@@ -0,0 +1,275 @@
+"""
+Local file transcription processor using faster-whisper with Silero VAD pipeline.
+
+Downloads audio from URL, segments it using Silero VAD, transcribes each
+segment with faster-whisper, and merges results. No HTTP backend needed.
+
+VAD pipeline ported from gpu/self_hosted/app/services/transcriber.py.
+"""
+
+import asyncio
+import os
+import shutil
+import subprocess
+import threading
+from typing import Generator
+
+import numpy as np
+from silero_vad import VADIterator, load_silero_vad
+
+from reflector.processors._audio_download import download_audio_to_temp
+from reflector.processors.file_transcript import (
+    FileTranscriptInput,
+    FileTranscriptProcessor,
+)
+from reflector.processors.file_transcript_auto import FileTranscriptAutoProcessor
+from reflector.processors.types import Transcript, Word
+from reflector.settings import settings
+
+SAMPLE_RATE = 16000
+
+VAD_CONFIG = {
+    "batch_max_duration": 30.0,
+    "silence_padding": 0.5,
+    "window_size": 512,
+}
+
+
+class FileTranscriptWhisperProcessor(FileTranscriptProcessor):
+    """Transcribe complete audio files using local faster-whisper with VAD."""
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self._model = None
+        self._lock = threading.Lock()
+
+    def _ensure_model(self):
+        """Lazy-load the whisper model on first use."""
+        if self._model is not None:
+            return
+
+        import faster_whisper
+        import torch
+
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        compute_type = "float16" if device == "cuda" else "int8"
+        model_name = settings.WHISPER_FILE_MODEL
+
+        self.logger.info(
+            "Loading whisper model",
+            model=model_name,
+            device=device,
+            compute_type=compute_type,
+        )
+        self._model = faster_whisper.WhisperModel(
+            model_name,
+            device=device,
+            compute_type=compute_type,
+            num_workers=1,
+        )
+
+    async def _transcript(self, data: FileTranscriptInput):
+        """Download file, run VAD segmentation, transcribe each segment."""
+        tmp_path = await download_audio_to_temp(data.audio_url)
+        try:
+            loop = asyncio.get_event_loop()
+            result = await loop.run_in_executor(
+                None,
+                self._transcribe_file_blocking,
+                str(tmp_path),
+                data.language,
+            )
+            return result
+        finally:
+            try:
+                os.unlink(tmp_path)
+            except OSError:
+                pass
+
+    def _transcribe_file_blocking(self, file_path: str, language: str) -> Transcript:
+        """Blocking transcription with VAD pipeline."""
+        self._ensure_model()
+
+        audio_array = _load_audio_via_ffmpeg(file_path, SAMPLE_RATE)
+
+        # VAD segmentation → batch merging
+        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))
+
+        # If no speech detected, try transcribing the whole file
+        if not merged_batches:
+            return self._transcribe_whole_file(file_path, language)
+
+        # Transcribe each batch
+        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 = _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)
+
+            for seg in segments:
+                for w in seg.words:
+                    all_words.append(
+                        {
+                            "word": w.word,
+                            "start": round(float(w.start) + start_time, 2),
+                            "end": round(float(w.end) + start_time, 2),
+                        }
+                    )
+
+        all_words = _enforce_word_timing_constraints(all_words)
+
+        words = [
+            Word(text=w["word"], start=w["start"], end=w["end"]) for w in all_words
+        ]
+        words.sort(key=lambda w: w.start)
+        return Transcript(words=words)
+
+    def _transcribe_whole_file(self, file_path: str, language: str) -> Transcript:
+        """Fallback: transcribe entire file without VAD segmentation."""
+        with self._lock:
+            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)
+
+        words = []
+        for seg in segments:
+            for w in seg.words:
+                words.append(
+                    Word(
+                        text=w.word,
+                        start=round(float(w.start), 2),
+                        end=round(float(w.end), 2),
+                    )
+                )
+        return Transcript(words=words)
+
+
+# --- VAD helpers (ported from gpu/self_hosted/app/services/transcriber.py) ---
+# IMPORTANT: This VAD segment logic is duplicated for deployment isolation.
+# If you modify this, consider updating the GPU service copy as well:
+#   - gpu/self_hosted/app/services/transcriber.py
+#   - gpu/modal_deployments/reflector_transcriber.py
+#   - gpu/modal_deployments/reflector_transcriber_parakeet.py
+
+
+def _load_audio_via_ffmpeg(
+    input_path: str, sample_rate: int = SAMPLE_RATE
+) -> np.ndarray:
+    """Load audio file via ffmpeg, converting to mono float32 at target sample rate."""
+    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",
+    ]
+    proc = subprocess.run(
+        cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE, check=True
+    )
+    return np.frombuffer(proc.stdout, dtype=np.float32)
+
+
+def _vad_segments(
+    audio_array: np.ndarray,
+    sample_rate: int = SAMPLE_RATE,
+    window_size: int = VAD_CONFIG["window_size"],
+) -> Generator[tuple[float, float], None, None]:
+    """Detect speech segments using Silero VAD."""
+    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
+
+    # Handle case where audio ends while speech is still active
+    if start is not None:
+        audio_duration = len(audio_array) / float(sample_rate)
+        yield (start / float(SAMPLE_RATE), audio_duration)
+
+    iterator.reset_states()
+
+
+def _pad_audio(audio_array: np.ndarray, sample_rate: int = SAMPLE_RATE) -> np.ndarray:
+    """Pad short audio with silence for VAD compatibility."""
+    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(words: list[dict]) -> list[dict]:
+    """Ensure no word end time exceeds the next word's start time."""
+    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
+
+
+FileTranscriptAutoProcessor.register("whisper", FileTranscriptWhisperProcessor)

server/reflector/processors/transcript_translator_marian.py

@@ -0,0 +1,50 @@
+"""
+MarianMT transcript translator processor using HuggingFace MarianMT in-process.
+
+Translates transcript text using HuggingFace MarianMT models
+locally. No HTTP backend needed.
+"""
+
+import asyncio
+
+from reflector.processors._marian_translator_service import translator_service
+from reflector.processors.transcript_translator import TranscriptTranslatorProcessor
+from reflector.processors.transcript_translator_auto import (
+    TranscriptTranslatorAutoProcessor,
+)
+from reflector.processors.types import TranslationLanguages
+
+
+class TranscriptTranslatorMarianProcessor(TranscriptTranslatorProcessor):
+    """Translate transcript text using MarianMT models."""
+
+    async def _translate(self, text: str) -> str | None:
+        source_language = self.get_pref("audio:source_language", "en")
+        target_language = self.get_pref("audio:target_language", "en")
+
+        languages = TranslationLanguages()
+        assert languages.is_supported(target_language)
+
+        self.logger.debug(f"MarianMT translate {text=}")
+
+        loop = asyncio.get_event_loop()
+        result = await loop.run_in_executor(
+            None,
+            translator_service.translate,
+            text,
+            source_language,
+            target_language,
+        )
+
+        if target_language in result["text"]:
+            translation = result["text"][target_language]
+        else:
+            translation = None
+
+        self.logger.debug(f"Translation result: {text=}, {translation=}")
+        return translation
+
+
+TranscriptTranslatorAutoProcessor.register(
+    "marian", TranscriptTranslatorMarianProcessor
+)

server/reflector/settings.py

@@ -40,14 +40,24 @@ class Settings(BaseSettings):
     # backends: silero, frames
     AUDIO_CHUNKER_BACKEND: str = "frames"
 
+    # HuggingFace token for gated models (pyannote diarization in --cpu mode)
+    HF_TOKEN: str | None = None
+
     # Audio Transcription
     # backends:
     #   - whisper: in-process model loading (no HTTP, runs in same process)
     #   - modal: HTTP API client (works with Modal.com OR self-hosted gpu/self_hosted/)
     TRANSCRIPT_BACKEND: str = "whisper"
+
+    # Whisper model sizes for local transcription
+    # Options: "tiny", "base", "small", "medium", "large-v2"
+    WHISPER_CHUNK_MODEL: str = "tiny"
+    WHISPER_FILE_MODEL: str = "tiny"
     TRANSCRIPT_URL: str | None = None
     TRANSCRIPT_TIMEOUT: int = 90
-    TRANSCRIPT_FILE_TIMEOUT: int = 600
+    TRANSCRIPT_FILE_TIMEOUT: int = (
+        540  # Below Hatchet TIMEOUT_HEAVY (600) to avoid timeout race
+    )
 
     # Audio Transcription: modal backend
     TRANSCRIPT_MODAL_API_KEY: str | None = None
@@ -100,7 +110,7 @@ class Settings(BaseSettings):
     )
 
     # Diarization
-    # backend: modal — HTTP API client (works with Modal.com OR self-hosted gpu/self_hosted/)
+    # backends: modal — HTTP API client, pyannote — in-process pyannote.audio
     DIARIZATION_ENABLED: bool = True
     DIARIZATION_BACKEND: str = "modal"
     DIARIZATION_URL: str | None = None
@@ -111,9 +121,9 @@ class Settings(BaseSettings):
 
     # Audio Padding
     # backends:
-    #   - local: in-process PyAV padding (no HTTP, runs in same process)
+    #   - pyav: in-process PyAV padding (no HTTP, runs in same process)
     #   - modal: HTTP API client (works with Modal.com OR self-hosted gpu/self_hosted/)
-    PADDING_BACKEND: str = "local"
+    PADDING_BACKEND: str = "pyav"
     PADDING_URL: str | None = None
     PADDING_MODAL_API_KEY: str | None = None
 

server/reflector/utils/retry.py

@@ -30,6 +30,7 @@ def retry(fn):
             "retry_httpx_status_stop",
             (
                 401,  # auth issue
+                402,  # payment required / no credits — needs human action
                 404,  # not found
                 413,  # payload too large
                 418,  # teapot
@@ -58,8 +59,9 @@ def retry(fn):
                 result = await fn(*args, **kwargs)
                 if isinstance(result, Response):
                     result.raise_for_status()
-                if result:
+                # Return any result including falsy (e.g. "" from get_response);
-                    return result
+                # only retry on exception, not on empty string.
+                return result
             except HTTPStatusError as e:
                 retry_logger.exception(e)
                 status_code = e.response.status_code

server/reflector/views/meetings.py

@@ -89,14 +89,16 @@ class StartRecordingRequest(BaseModel):
 
 @router.post("/meetings/{meeting_id}/recordings/start")
 async def start_recording(
-    meeting_id: NonEmptyString, body: StartRecordingRequest
+    meeting_id: NonEmptyString,
+    body: StartRecordingRequest,
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
+    ],
 ) -> dict[str, Any]:
     """Start cloud or raw-tracks recording via Daily.co REST API.
 
     Both cloud and raw-tracks are started via REST API to bypass enable_recording limitation of allowing only 1 recording at a time.
     Uses different instanceIds for cloud vs raw-tracks (same won't work)
-
-    Note: No authentication required - anonymous users supported. TODO this is a DOS vector
     """
     meeting = await meetings_controller.get_by_id(meeting_id)
     if not meeting:

server/reflector/views/rooms.py

@@ -17,7 +17,6 @@ from reflector.db.rooms import rooms_controller
 from reflector.redis_cache import RedisAsyncLock
 from reflector.schemas.platform import Platform
 from reflector.services.ics_sync import ics_sync_service
-from reflector.settings import settings
 from reflector.utils.url import add_query_param
 from reflector.video_platforms.factory import create_platform_client
 from reflector.worker.webhook import test_webhook
@@ -178,11 +177,10 @@ router = APIRouter()
 
 @router.get("/rooms", response_model=Page[RoomDetails])
 async def rooms_list(
-    user: Annotated[Optional[auth.UserInfo], Depends(auth.current_user_optional)],
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
+    ],
 ) -> list[RoomDetails]:
-    if not user and not settings.PUBLIC_MODE:
-        raise HTTPException(status_code=401, detail="Not authenticated")
-
     user_id = user["sub"] if user else None
 
     paginated = await apaginate(

server/reflector/views/transcripts.py

@@ -263,16 +263,15 @@ class SearchResponse(BaseModel):
 
 @router.get("/transcripts", response_model=Page[GetTranscriptMinimal])
 async def transcripts_list(
-    user: Annotated[Optional[auth.UserInfo], Depends(auth.current_user_optional)],
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
+    ],
     source_kind: SourceKind | None = None,
     room_id: str | None = None,
     search_term: str | None = None,
     change_seq_from: int | None = None,
     sort_by: Literal["created_at", "change_seq"] | None = None,
 ):
-    if not user and not settings.PUBLIC_MODE:
-        raise HTTPException(status_code=401, detail="Not authenticated")
-
     user_id = user["sub"] if user else None
 
     # Default behavior preserved: sort_by=None → "-created_at"
@@ -307,13 +306,10 @@ async def transcripts_search(
     from_datetime: SearchFromDatetimeParam = None,
     to_datetime: SearchToDatetimeParam = None,
     user: Annotated[
-        Optional[auth.UserInfo], Depends(auth.current_user_optional)
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
     ] = None,
 ):
     """Full-text search across transcript titles and content."""
-    if not user and not settings.PUBLIC_MODE:
-        raise HTTPException(status_code=401, detail="Not authenticated")
-
     user_id = user["sub"] if user else None
 
     if from_datetime and to_datetime and from_datetime > to_datetime:
@@ -346,7 +342,9 @@ async def transcripts_search(
 @router.post("/transcripts", response_model=GetTranscriptWithParticipants)
 async def transcripts_create(
     info: CreateTranscript,
-    user: Annotated[Optional[auth.UserInfo], Depends(auth.current_user_optional)],
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
+    ],
 ):
     user_id = user["sub"] if user else None
     transcript = await transcripts_controller.add(

server/reflector/views/transcripts_audio.py

@@ -13,7 +13,6 @@ from fastapi import APIRouter, Depends, HTTPException, Request, Response, status
 import reflector.auth as auth
 from reflector.db.transcripts import AudioWaveform, transcripts_controller
 from reflector.settings import settings
-from reflector.views.transcripts import ALGORITHM
 
 from ._range_requests_response import range_requests_response
 
@@ -36,16 +35,23 @@ async def transcript_get_audio_mp3(
 ):
     user_id = user["sub"] if user else None
     if not user_id and token:
-        unauthorized_exception = HTTPException(
-            status_code=status.HTTP_401_UNAUTHORIZED,
-            detail="Invalid or expired token",
-            headers={"WWW-Authenticate": "Bearer"},
-        )
         try:
-            payload = jwt.decode(token, settings.SECRET_KEY, algorithms=[ALGORITHM])
+            token_user = await auth.verify_raw_token(token)
-            user_id: str = payload.get("sub")
+        except Exception:
-        except jwt.PyJWTError:
+            token_user = None
-            raise unauthorized_exception
+        # Fallback: try as internal HS256 token (created by _generate_local_audio_link)
+        if not token_user:
+            try:
+                payload = jwt.decode(token, settings.SECRET_KEY, algorithms=["HS256"])
+                user_id = payload.get("sub")
+            except jwt.PyJWTError:
+                raise HTTPException(
+                    status_code=status.HTTP_401_UNAUTHORIZED,
+                    detail="Invalid or expired token",
+                    headers={"WWW-Authenticate": "Bearer"},
+                )
+        else:
+            user_id = token_user["sub"]
 
     transcript = await transcripts_controller.get_by_id_for_http(
         transcript_id, user_id=user_id

server/reflector/views/transcripts_participants.py

@@ -62,8 +62,7 @@ async def transcript_add_participant(
     transcript = await transcripts_controller.get_by_id_for_http(
         transcript_id, user_id=user_id
     )
-    if transcript.user_id is not None and transcript.user_id != user_id:
+    transcripts_controller.check_can_mutate(transcript, user_id)
-        raise HTTPException(status_code=403, detail="Not authorized")
 
     # ensure the speaker is unique
     if participant.speaker is not None and transcript.participants is not None:
@@ -109,8 +108,7 @@ async def transcript_update_participant(
     transcript = await transcripts_controller.get_by_id_for_http(
         transcript_id, user_id=user_id
     )
-    if transcript.user_id is not None and transcript.user_id != user_id:
+    transcripts_controller.check_can_mutate(transcript, user_id)
-        raise HTTPException(status_code=403, detail="Not authorized")
 
     # ensure the speaker is unique
     for p in transcript.participants:
@@ -148,7 +146,6 @@ async def transcript_delete_participant(
     transcript = await transcripts_controller.get_by_id_for_http(
         transcript_id, user_id=user_id
     )
-    if transcript.user_id is not None and transcript.user_id != user_id:
+    transcripts_controller.check_can_mutate(transcript, user_id)
-        raise HTTPException(status_code=403, detail="Not authorized")
     await transcripts_controller.delete_participant(transcript, participant_id)
     return DeletionStatus(status="ok")

server/reflector/views/transcripts_process.py

@@ -26,7 +26,9 @@ class ProcessStatus(BaseModel):
 @router.post("/transcripts/{transcript_id}/process")
 async def transcript_process(
     transcript_id: str,
-    user: Annotated[Optional[auth.UserInfo], Depends(auth.current_user_optional)],
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
+    ],
 ) -> ProcessStatus:
     user_id = user["sub"] if user else None
     transcript = await transcripts_controller.get_by_id_for_http(
@@ -50,5 +52,8 @@ async def transcript_process(
     if isinstance(config, ProcessError):
         raise HTTPException(status_code=500, detail=config.detail)
     else:
-        await dispatch_transcript_processing(config)
+        # When transcript is in error state, force a new workflow instead of replaying
+        # (replay would re-run from failure point with same conditions and likely fail again)
+        force = transcript.status == "error"
+        await dispatch_transcript_processing(config, force=force)
         return ProcessStatus(status="ok")

server/reflector/views/transcripts_speaker.py

@@ -41,8 +41,7 @@ async def transcript_assign_speaker(
     transcript = await transcripts_controller.get_by_id_for_http(
         transcript_id, user_id=user_id
     )
-    if transcript.user_id is not None and transcript.user_id != user_id:
+    transcripts_controller.check_can_mutate(transcript, user_id)
-        raise HTTPException(status_code=403, detail="Not authorized")
 
     if not transcript:
         raise HTTPException(status_code=404, detail="Transcript not found")
@@ -121,8 +120,7 @@ async def transcript_merge_speaker(
     transcript = await transcripts_controller.get_by_id_for_http(
         transcript_id, user_id=user_id
     )
-    if transcript.user_id is not None and transcript.user_id != user_id:
+    transcripts_controller.check_can_mutate(transcript, user_id)
-        raise HTTPException(status_code=403, detail="Not authorized")
 
     if not transcript:
         raise HTTPException(status_code=404, detail="Transcript not found")

server/reflector/views/transcripts_upload.py

@@ -21,7 +21,9 @@ async def transcript_record_upload(
     chunk_number: int,
     total_chunks: int,
     chunk: UploadFile,
-    user: Annotated[Optional[auth.UserInfo], Depends(auth.current_user_optional)],
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
+    ],
 ):
     user_id = user["sub"] if user else None
     transcript = await transcripts_controller.get_by_id_for_http(

server/reflector/views/transcripts_webrtc.py

@@ -15,7 +15,9 @@ async def transcript_record_webrtc(
     transcript_id: str,
     params: RtcOffer,
     request: Request,
-    user: Annotated[Optional[auth.UserInfo], Depends(auth.current_user_optional)],
+    user: Annotated[
+        Optional[auth.UserInfo], Depends(auth.current_user_optional_if_public_mode)
+    ],
 ):
     user_id = user["sub"] if user else None
     transcript = await transcripts_controller.get_by_id_for_http(

server/reflector/worker/process.py

@@ -132,7 +132,7 @@ async def process_recording(bucket_name: str, object_key: str):
             target_language="en",
             user_id=room.user_id,
             recording_id=recording.id,
-            share_mode="public",
+            share_mode="semi-private",
             meeting_id=meeting.id,
             room_id=room.id,
         )
@@ -343,7 +343,7 @@ async def _process_multitrack_recording_inner(
             target_language="en",
             user_id=room.user_id,
             recording_id=recording.id,
-            share_mode="public",
+            share_mode="semi-private",
             meeting_id=meeting.id,
             room_id=room.id,
         )

server/tests/conftest.py

@@ -437,6 +437,8 @@ async def ws_manager_in_memory(monkeypatch):
 
     try:
         fastapi_app.dependency_overrides[auth.current_user_optional] = lambda: None
+        # current_user_optional_if_public_mode is NOT overridden here so the real
+        # implementation runs and enforces the PUBLIC_MODE check during tests.
     except Exception:
         pass
 
@@ -491,37 +493,39 @@ async def authenticated_client2():
 @asynccontextmanager
 async def authenticated_client_ctx():
     from reflector.app import app
-    from reflector.auth import current_user, current_user_optional
+    from reflector.auth import (
+        current_user,
+        current_user_optional,
+        current_user_optional_if_public_mode,
+    )
 
-    app.dependency_overrides[current_user] = lambda: {
+    _user = lambda: {"sub": "randomuserid", "email": "test@mail.com"}
-        "sub": "randomuserid",
+    app.dependency_overrides[current_user] = _user
-        "email": "test@mail.com",
+    app.dependency_overrides[current_user_optional] = _user
-    }
+    app.dependency_overrides[current_user_optional_if_public_mode] = _user
-    app.dependency_overrides[current_user_optional] = lambda: {
-        "sub": "randomuserid",
-        "email": "test@mail.com",
-    }
     yield
     del app.dependency_overrides[current_user]
     del app.dependency_overrides[current_user_optional]
+    del app.dependency_overrides[current_user_optional_if_public_mode]
 
 
 @asynccontextmanager
 async def authenticated_client2_ctx():
     from reflector.app import app
-    from reflector.auth import current_user, current_user_optional
+    from reflector.auth import (
+        current_user,
+        current_user_optional,
+        current_user_optional_if_public_mode,
+    )
 
-    app.dependency_overrides[current_user] = lambda: {
+    _user = lambda: {"sub": "randomuserid2", "email": "test@mail.com"}
-        "sub": "randomuserid2",
+    app.dependency_overrides[current_user] = _user
-        "email": "test@mail.com",
+    app.dependency_overrides[current_user_optional] = _user
-    }
+    app.dependency_overrides[current_user_optional_if_public_mode] = _user
-    app.dependency_overrides[current_user_optional] = lambda: {
-        "sub": "randomuserid2",
-        "email": "test@mail.com",
-    }
     yield
     del app.dependency_overrides[current_user]
     del app.dependency_overrides[current_user_optional]
+    del app.dependency_overrides[current_user_optional_if_public_mode]
 
 
 @pytest.fixture(scope="session")
@@ -550,7 +554,7 @@ def reset_hatchet_client():
 
 
 @pytest.fixture
-async def fake_transcript_with_topics(tmpdir, client):
+async def fake_transcript_with_topics(tmpdir, client, monkeypatch):
     import shutil
     from pathlib import Path
 
@@ -559,6 +563,9 @@ async def fake_transcript_with_topics(tmpdir, client):
     from reflector.settings import settings
     from reflector.views.transcripts import transcripts_controller
 
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     settings.DATA_DIR = Path(tmpdir)
 
     # create a transcript

server/tests/test_app.py

@@ -0,0 +1,17 @@
+"""Tests for app-level endpoints (root, not under /v1)."""
+
+import pytest
+
+
+@pytest.mark.asyncio
+async def test_health_endpoint_returns_healthy():
+    """GET /health returns 200 and {"status": "healthy"} for probes and CI."""
+    from httpx import AsyncClient
+
+    from reflector.app import app
+
+    # Health is at app root, not under /v1
+    async with AsyncClient(app=app, base_url="http://test") as root_client:
+        response = await root_client.get("/health")
+    assert response.status_code == 200
+    assert response.json() == {"status": "healthy"}

server/tests/test_hatchet_error_handling.py

@@ -0,0 +1,303 @@
+"""
+Tests for Hatchet error handling: NonRetryable classification and error status.
+
+These tests encode the desired behavior from the Hatchet Workflow Analysis doc:
+- Transient exceptions: do NOT set error status (let Hatchet retry; user stays on "processing").
+- Hard-fail exceptions: set error status and re-raise as NonRetryableException (stop retries).
+- on_failure_task: sets error status when workflow is truly dead.
+
+Run before the fix: some tests fail (reproducing the issues).
+Run after the fix: all tests pass.
+"""
+
+from contextlib import asynccontextmanager
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import httpx
+import pytest
+from hatchet_sdk import NonRetryableException
+
+from reflector.hatchet.error_classification import is_non_retryable
+from reflector.llm import LLMParseError
+
+# --- Tests for is_non_retryable() (pass once error_classification exists) ---
+
+
+def test_is_non_retryable_returns_true_for_value_error():
+    """ValueError (e.g. missing config) should stop retries."""
+    assert is_non_retryable(ValueError("DAILY_API_KEY must be set")) is True
+
+
+def test_is_non_retryable_returns_true_for_type_error():
+    """TypeError (bad input) should stop retries."""
+    assert is_non_retryable(TypeError("expected str")) is True
+
+
+def test_is_non_retryable_returns_true_for_http_401():
+    """HTTP 401 auth error should stop retries."""
+    resp = MagicMock()
+    resp.status_code = 401
+    err = httpx.HTTPStatusError("Unauthorized", request=MagicMock(), response=resp)
+    assert is_non_retryable(err) is True
+
+
+def test_is_non_retryable_returns_true_for_http_402():
+    """HTTP 402 (no credits) should stop retries."""
+    resp = MagicMock()
+    resp.status_code = 402
+    err = httpx.HTTPStatusError("Payment Required", request=MagicMock(), response=resp)
+    assert is_non_retryable(err) is True
+
+
+def test_is_non_retryable_returns_true_for_http_404():
+    """HTTP 404 should stop retries."""
+    resp = MagicMock()
+    resp.status_code = 404
+    err = httpx.HTTPStatusError("Not Found", request=MagicMock(), response=resp)
+    assert is_non_retryable(err) is True
+
+
+def test_is_non_retryable_returns_false_for_http_503():
+    """HTTP 503 is transient; retries are useful."""
+    resp = MagicMock()
+    resp.status_code = 503
+    err = httpx.HTTPStatusError(
+        "Service Unavailable", request=MagicMock(), response=resp
+    )
+    assert is_non_retryable(err) is False
+
+
+def test_is_non_retryable_returns_false_for_timeout():
+    """Timeout is transient."""
+    assert is_non_retryable(httpx.TimeoutException("timed out")) is False
+
+
+def test_is_non_retryable_returns_true_for_llm_parse_error():
+    """LLMParseError after internal retries should stop."""
+    from pydantic import BaseModel
+
+    class _Dummy(BaseModel):
+        pass
+
+    assert is_non_retryable(LLMParseError(_Dummy, "Failed to parse", 3)) is True
+
+
+def test_is_non_retryable_returns_true_for_non_retryable_exception():
+    """Already-wrapped NonRetryableException should stay non-retryable."""
+    assert is_non_retryable(NonRetryableException("custom")) is True
+
+
+# --- Tests for with_error_handling (need pipeline module with patch) ---
+
+
+@pytest.fixture(scope="module")
+def pipeline_module():
+    """Import daily_multitrack_pipeline with Hatchet client mocked."""
+    with patch("reflector.hatchet.client.settings") as s:
+        s.HATCHET_CLIENT_TOKEN = "test-token"
+        s.HATCHET_DEBUG = False
+    mock_client = MagicMock()
+    mock_client.workflow.return_value = MagicMock()
+    with patch(
+        "reflector.hatchet.client.HatchetClientManager.get_client",
+        return_value=mock_client,
+    ):
+        from reflector.hatchet.workflows import daily_multitrack_pipeline
+
+        return daily_multitrack_pipeline
+
+
+@pytest.fixture
+def mock_input():
+    """Minimal PipelineInput for decorator tests."""
+    from reflector.hatchet.workflows.daily_multitrack_pipeline import PipelineInput
+
+    return PipelineInput(
+        recording_id="rec-1",
+        tracks=[],
+        bucket_name="bucket",
+        transcript_id="ts-123",
+        room_id=None,
+    )
+
+
+@pytest.fixture
+def mock_ctx():
+    """Minimal Context-like object."""
+    ctx = MagicMock()
+    ctx.log = MagicMock()
+    return ctx
+
+
+@pytest.mark.asyncio
+async def test_with_error_handling_transient_does_not_set_error_status(
+    pipeline_module, mock_input, mock_ctx
+):
+    """Transient exception must NOT set error status (so user stays on 'processing' during retries).
+
+    Before fix: set_workflow_error_status is called on every exception → FAIL.
+    After fix: not called for transient → PASS.
+    """
+    from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+        TaskName,
+        with_error_handling,
+    )
+
+    async def failing_task(input, ctx):
+        raise httpx.TimeoutException("timed out")
+
+    wrapped = with_error_handling(TaskName.GET_RECORDING)(failing_task)
+
+    with patch(
+        "reflector.hatchet.workflows.daily_multitrack_pipeline.set_workflow_error_status",
+        new_callable=AsyncMock,
+    ) as mock_set_error:
+        with pytest.raises(httpx.TimeoutException):
+            await wrapped(mock_input, mock_ctx)
+
+        # Desired: do NOT set error status for transient (Hatchet will retry)
+        mock_set_error.assert_not_called()
+
+
+@pytest.mark.asyncio
+async def test_with_error_handling_hard_fail_raises_non_retryable_and_sets_status(
+    pipeline_module, mock_input, mock_ctx
+):
+    """Hard-fail (e.g. ValueError) must set error status and re-raise NonRetryableException.
+
+    Before fix: raises ValueError, set_workflow_error_status called → test would need to expect ValueError.
+    After fix: raises NonRetryableException, set_workflow_error_status called → PASS.
+    """
+    from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+        TaskName,
+        with_error_handling,
+    )
+
+    async def failing_task(input, ctx):
+        raise ValueError("PADDING_URL must be set")
+
+    wrapped = with_error_handling(TaskName.GET_RECORDING)(failing_task)
+
+    with patch(
+        "reflector.hatchet.workflows.daily_multitrack_pipeline.set_workflow_error_status",
+        new_callable=AsyncMock,
+    ) as mock_set_error:
+        with pytest.raises(NonRetryableException) as exc_info:
+            await wrapped(mock_input, mock_ctx)
+
+        assert "PADDING_URL" in str(exc_info.value)
+        mock_set_error.assert_called_once_with("ts-123")
+
+
+@pytest.mark.asyncio
+async def test_with_error_handling_set_error_status_false_never_sets_status(
+    pipeline_module, mock_input, mock_ctx
+):
+    """When set_error_status=False, we must never set error status (e.g. cleanup_consent)."""
+    from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+        TaskName,
+        with_error_handling,
+    )
+
+    async def failing_task(input, ctx):
+        raise ValueError("something went wrong")
+
+    wrapped = with_error_handling(TaskName.CLEANUP_CONSENT, set_error_status=False)(
+        failing_task
+    )
+
+    with patch(
+        "reflector.hatchet.workflows.daily_multitrack_pipeline.set_workflow_error_status",
+        new_callable=AsyncMock,
+    ) as mock_set_error:
+        with pytest.raises((ValueError, NonRetryableException)):
+            await wrapped(mock_input, mock_ctx)
+
+        mock_set_error.assert_not_called()
+
+
+@asynccontextmanager
+async def _noop_db_context():
+    """Async context manager that yields without touching the DB (for unit tests)."""
+    yield None
+
+
+@pytest.mark.asyncio
+async def test_on_failure_task_sets_error_status(pipeline_module, mock_input, mock_ctx):
+    """When workflow fails and transcript is not yet 'ended', on_failure sets status to 'error'."""
+    from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+        on_workflow_failure,
+    )
+
+    transcript_processing = MagicMock()
+    transcript_processing.status = "processing"
+
+    with patch(
+        "reflector.hatchet.workflows.daily_multitrack_pipeline.fresh_db_connection",
+        _noop_db_context,
+    ):
+        with patch(
+            "reflector.db.transcripts.transcripts_controller.get_by_id",
+            new_callable=AsyncMock,
+            return_value=transcript_processing,
+        ):
+            with patch(
+                "reflector.hatchet.workflows.daily_multitrack_pipeline.set_workflow_error_status",
+                new_callable=AsyncMock,
+            ) as mock_set_error:
+                await on_workflow_failure(mock_input, mock_ctx)
+                mock_set_error.assert_called_once_with(mock_input.transcript_id)
+
+
+@pytest.mark.asyncio
+async def test_on_failure_task_does_not_overwrite_ended(
+    pipeline_module, mock_input, mock_ctx
+):
+    """When workflow fails after finalize (e.g. post_zulip), do not overwrite 'ended' with 'error'.
+
+    cleanup_consent, post_zulip, send_webhook use set_error_status=False; if one fails,
+    on_workflow_failure must not set status to 'error' when transcript is already 'ended'.
+    """
+    from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+        on_workflow_failure,
+    )
+
+    transcript_ended = MagicMock()
+    transcript_ended.status = "ended"
+
+    with patch(
+        "reflector.hatchet.workflows.daily_multitrack_pipeline.fresh_db_connection",
+        _noop_db_context,
+    ):
+        with patch(
+            "reflector.db.transcripts.transcripts_controller.get_by_id",
+            new_callable=AsyncMock,
+            return_value=transcript_ended,
+        ):
+            with patch(
+                "reflector.hatchet.workflows.daily_multitrack_pipeline.set_workflow_error_status",
+                new_callable=AsyncMock,
+            ) as mock_set_error:
+                await on_workflow_failure(mock_input, mock_ctx)
+                mock_set_error.assert_not_called()
+
+
+# --- Tests for fan-out helper (_successful_run_results) ---
+
+
+def test_successful_run_results_filters_exceptions():
+    """_successful_run_results returns only non-exception items from aio_run_many(return_exceptions=True)."""
+    from reflector.hatchet.workflows.daily_multitrack_pipeline import (
+        _successful_run_results,
+    )
+
+    results = [
+        {"key": "ok1"},
+        ValueError("child failed"),
+        {"key": "ok2"},
+        RuntimeError("another"),
+    ]
+    successful = _successful_run_results(results)
+    assert len(successful) == 2
+    assert successful[0] == {"key": "ok1"}
+    assert successful[1] == {"key": "ok2"}

server/tests/test_llm_retry.py

@@ -1,6 +1,6 @@
 """Tests for LLM structured output with astructured_predict + reflection retry"""
 
-from unittest.mock import AsyncMock, patch
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 from pydantic import BaseModel, Field, ValidationError
@@ -252,6 +252,63 @@ class TestNetworkErrorRetries:
             assert mock_settings.llm.astructured_predict.call_count == 3
 
 
+class TestGetResponseRetries:
+    """Test that get_response() uses the same retry() wrapper for transient errors."""
+
+    @pytest.mark.asyncio
+    async def test_get_response_retries_on_connection_error(self, test_settings):
+        """Test that get_response retries on ConnectionError and returns on success."""
+        llm = LLM(settings=test_settings, temperature=0.4, max_tokens=100)
+
+        mock_instance = MagicMock()
+        mock_instance.aget_response = AsyncMock(
+            side_effect=[
+                ConnectionError("Connection refused"),
+                "  Summary text  ",
+            ]
+        )
+
+        with patch("reflector.llm.TreeSummarize", return_value=mock_instance):
+            result = await llm.get_response("Prompt", ["text"])
+
+        assert result == "Summary text"
+        assert mock_instance.aget_response.call_count == 2
+
+    @pytest.mark.asyncio
+    async def test_get_response_exhausts_retries(self, test_settings):
+        """Test that get_response raises RetryException after retry attempts exceeded."""
+        llm = LLM(settings=test_settings, temperature=0.4, max_tokens=100)
+
+        mock_instance = MagicMock()
+        mock_instance.aget_response = AsyncMock(
+            side_effect=ConnectionError("Connection refused")
+        )
+
+        with patch("reflector.llm.TreeSummarize", return_value=mock_instance):
+            with pytest.raises(RetryException, match="Retry attempts exceeded"):
+                await llm.get_response("Prompt", ["text"])
+
+        assert mock_instance.aget_response.call_count == 3
+
+    @pytest.mark.asyncio
+    async def test_get_response_returns_empty_string_without_retry(self, test_settings):
+        """Empty or whitespace-only LLM response must return '' and not raise RetryException.
+
+        retry() must return falsy results (e.g. '' from get_response) instead of
+        treating them as 'no result' and retrying until RetryException.
+        """
+        llm = LLM(settings=test_settings, temperature=0.4, max_tokens=100)
+
+        mock_instance = MagicMock()
+        mock_instance.aget_response = AsyncMock(return_value="   \n  ")  # strip() -> ""
+
+        with patch("reflector.llm.TreeSummarize", return_value=mock_instance):
+            result = await llm.get_response("Prompt", ["text"])
+
+        assert result == ""
+        assert mock_instance.aget_response.call_count == 1
+
+
 class TestTextsInclusion:
     """Test that texts parameter is included in the prompt sent to astructured_predict"""
 

server/tests/test_pipeline_main_file.py

@@ -5,6 +5,7 @@ This test verifies the complete file processing pipeline without mocking much,
 ensuring all processors are correctly invoked and the happy path works correctly.
 """
 
+from contextlib import asynccontextmanager
 from pathlib import Path
 from unittest.mock import AsyncMock, MagicMock, patch
 from uuid import uuid4
@@ -651,3 +652,43 @@ async def test_pipeline_file_process_no_audio_file(
     # This should fail when trying to open the file with av
     with pytest.raises(Exception):
         await pipeline.process(non_existent_path)
+
+
+@pytest.mark.asyncio
+async def test_on_title_does_not_overwrite_user_set_title():
+    """When transcript already has a title, on_title does not call update."""
+    from reflector.db.transcripts import Transcript, TranscriptFinalTitle
+    from reflector.pipelines.main_file_pipeline import PipelineMainFile
+
+    transcript_id = str(uuid4())
+    transcript_with_title = Transcript(
+        id=transcript_id,
+        name="test",
+        source_kind="file",
+        title="User set title",
+    )
+
+    controller = "reflector.pipelines.main_live_pipeline.transcripts_controller"
+    with patch(f"{controller}.get_by_id", new_callable=AsyncMock) as mock_get:
+        with patch(f"{controller}.update", new_callable=AsyncMock) as mock_update:
+            with patch(
+                f"{controller}.append_event", new_callable=AsyncMock
+            ) as mock_append:
+                with patch(f"{controller}.transaction") as mock_txn:
+                    mock_get.return_value = transcript_with_title
+                    mock_append.return_value = None
+
+                    @asynccontextmanager
+                    async def noop_txn():
+                        yield
+
+                    mock_txn.return_value = noop_txn()
+
+                    pipeline = PipelineMainFile(transcript_id=transcript_id)
+                    await pipeline.on_title(
+                        TranscriptFinalTitle(title="Generated title")
+                    )
+
+                    mock_get.assert_called_once()
+                    mock_update.assert_not_called()
+                    mock_append.assert_called_once()

server/tests/test_processors_cpu.py

@@ -0,0 +1,450 @@
+"""
+Tests for in-process processor backends (--cpu mode).
+
+All ML model calls are mocked — no actual model loading needed.
+Tests verify processor registration, wiring, error handling, and data flow.
+"""
+
+import os
+import tempfile
+from pathlib import Path
+from unittest.mock import AsyncMock, MagicMock, patch
+
+import pytest
+
+from reflector.processors.file_diarization import (
+    FileDiarizationInput,
+    FileDiarizationOutput,
+)
+from reflector.processors.types import (
+    AudioDiarizationInput,
+    TitleSummaryWithId,
+    Transcript,
+    Word,
+)
+
+# ── Registration Tests ──────────────────────────────────────────────────
+
+
+def test_audio_diarization_pyannote_registers():
+    """Verify AudioDiarizationPyannoteProcessor registers with 'pyannote' backend."""
+    # Importing the module triggers registration
+    import reflector.processors.audio_diarization_pyannote  # noqa: F401
+    from reflector.processors.audio_diarization_auto import (
+        AudioDiarizationAutoProcessor,
+    )
+
+    assert "pyannote" in AudioDiarizationAutoProcessor._registry
+
+
+def test_file_diarization_pyannote_registers():
+    """Verify FileDiarizationPyannoteProcessor registers with 'pyannote' backend."""
+    import reflector.processors.file_diarization_pyannote  # noqa: F401
+    from reflector.processors.file_diarization_auto import FileDiarizationAutoProcessor
+
+    assert "pyannote" in FileDiarizationAutoProcessor._registry
+
+
+def test_transcript_translator_marian_registers():
+    """Verify TranscriptTranslatorMarianProcessor registers with 'marian' backend."""
+    import reflector.processors.transcript_translator_marian  # noqa: F401
+    from reflector.processors.transcript_translator_auto import (
+        TranscriptTranslatorAutoProcessor,
+    )
+
+    assert "marian" in TranscriptTranslatorAutoProcessor._registry
+
+
+def test_file_transcript_whisper_registers():
+    """Verify FileTranscriptWhisperProcessor registers with 'whisper' backend."""
+    import reflector.processors.file_transcript_whisper  # noqa: F401
+    from reflector.processors.file_transcript_auto import FileTranscriptAutoProcessor
+
+    assert "whisper" in FileTranscriptAutoProcessor._registry
+
+
+# ── Audio Download Utility Tests ────────────────────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_download_audio_to_temp_success():
+    """Verify download_audio_to_temp downloads to a temp file and returns path."""
+    from reflector.processors._audio_download import download_audio_to_temp
+
+    mock_response = MagicMock()
+    mock_response.status_code = 200
+    mock_response.headers = {"content-type": "audio/wav"}
+    mock_response.iter_content.return_value = [b"fake audio data"]
+    mock_response.raise_for_status = MagicMock()
+
+    with patch("reflector.processors._audio_download.requests.get") as mock_get:
+        mock_get.return_value = mock_response
+
+        result = await download_audio_to_temp("https://example.com/test.wav")
+
+        assert isinstance(result, Path)
+        assert result.exists()
+        assert result.read_bytes() == b"fake audio data"
+        assert result.suffix == ".wav"
+
+        # Cleanup
+        os.unlink(result)
+
+
+@pytest.mark.asyncio
+async def test_download_audio_to_temp_cleanup_on_error():
+    """Verify temp file is cleaned up when download fails mid-write."""
+    from reflector.processors._audio_download import download_audio_to_temp
+
+    mock_response = MagicMock()
+    mock_response.headers = {"content-type": "audio/wav"}
+    mock_response.raise_for_status = MagicMock()
+
+    def fail_iter(*args, **kwargs):
+        raise ConnectionError("Download interrupted")
+
+    mock_response.iter_content = fail_iter
+
+    with patch("reflector.processors._audio_download.requests.get") as mock_get:
+        mock_get.return_value = mock_response
+
+        with pytest.raises(ConnectionError, match="Download interrupted"):
+            await download_audio_to_temp("https://example.com/test.wav")
+
+
+def test_detect_extension_from_url():
+    """Verify extension detection from URL path."""
+    from reflector.processors._audio_download import _detect_extension
+
+    assert _detect_extension("https://example.com/test.wav", "") == ".wav"
+    assert _detect_extension("https://example.com/test.mp3?signed=1", "") == ".mp3"
+    assert _detect_extension("https://example.com/test.webm", "") == ".webm"
+
+
+def test_detect_extension_from_content_type():
+    """Verify extension detection from content-type header."""
+    from reflector.processors._audio_download import _detect_extension
+
+    assert _detect_extension("https://s3.aws/uuid", "audio/mpeg") == ".mp3"
+    assert _detect_extension("https://s3.aws/uuid", "audio/wav") == ".wav"
+    assert _detect_extension("https://s3.aws/uuid", "audio/webm") == ".webm"
+
+
+def test_detect_extension_fallback():
+    """Verify fallback extension when neither URL nor content-type is recognized."""
+    from reflector.processors._audio_download import _detect_extension
+
+    assert (
+        _detect_extension("https://s3.aws/uuid", "application/octet-stream") == ".audio"
+    )
+
+
+# ── Audio Diarization Pyannote Processor Tests ──────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_audio_diarization_pyannote_diarize():
+    """Verify pyannote audio diarization downloads, diarizes, and cleans up."""
+    from reflector.processors.audio_diarization_pyannote import (
+        AudioDiarizationPyannoteProcessor,
+    )
+
+    mock_diarization_result = {
+        "diarization": [
+            {"start": 0.0, "end": 2.5, "speaker": 0},
+            {"start": 2.5, "end": 5.0, "speaker": 1},
+        ]
+    }
+
+    # Create a temp file to simulate download
+    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
+    tmp.write(b"fake audio")
+    tmp.close()
+    tmp_path = Path(tmp.name)
+
+    processor = AudioDiarizationPyannoteProcessor()
+
+    with (
+        patch(
+            "reflector.processors.audio_diarization_pyannote.download_audio_to_temp",
+            new_callable=AsyncMock,
+            return_value=tmp_path,
+        ),
+        patch(
+            "reflector.processors.audio_diarization_pyannote.diarization_service"
+        ) as mock_svc,
+    ):
+        mock_svc.diarize_file.return_value = mock_diarization_result
+
+        data = AudioDiarizationInput(
+            audio_url="https://example.com/test.wav",
+            topics=[
+                TitleSummaryWithId(
+                    id="topic-1",
+                    title="Test Topic",
+                    summary="A test topic",
+                    timestamp=0.0,
+                    duration=5.0,
+                    transcript=Transcript(
+                        words=[Word(text="hello", start=0.0, end=1.0)]
+                    ),
+                )
+            ],
+        )
+        result = await processor._diarize(data)
+
+        assert result == mock_diarization_result["diarization"]
+        mock_svc.diarize_file.assert_called_once()
+
+
+# ── File Diarization Pyannote Processor Tests ───────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_file_diarization_pyannote_diarize():
+    """Verify pyannote file diarization returns FileDiarizationOutput."""
+    from reflector.processors.file_diarization_pyannote import (
+        FileDiarizationPyannoteProcessor,
+    )
+
+    mock_diarization_result = {
+        "diarization": [
+            {"start": 0.0, "end": 3.0, "speaker": 0},
+            {"start": 3.0, "end": 6.0, "speaker": 1},
+        ]
+    }
+
+    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
+    tmp.write(b"fake audio")
+    tmp.close()
+    tmp_path = Path(tmp.name)
+
+    processor = FileDiarizationPyannoteProcessor()
+
+    with (
+        patch(
+            "reflector.processors.file_diarization_pyannote.download_audio_to_temp",
+            new_callable=AsyncMock,
+            return_value=tmp_path,
+        ),
+        patch(
+            "reflector.processors.file_diarization_pyannote.diarization_service"
+        ) as mock_svc,
+    ):
+        mock_svc.diarize_file.return_value = mock_diarization_result
+
+        data = FileDiarizationInput(audio_url="https://example.com/test.wav")
+        result = await processor._diarize(data)
+
+        assert isinstance(result, FileDiarizationOutput)
+        assert len(result.diarization) == 2
+        assert result.diarization[0]["start"] == 0.0
+        assert result.diarization[1]["speaker"] == 1
+
+
+# ── Transcript Translator Marian Processor Tests ───────────────────────
+
+
+@pytest.mark.asyncio
+async def test_transcript_translator_marian_translate():
+    """Verify MarianMT translator calls service and extracts translation."""
+    from reflector.processors.transcript_translator_marian import (
+        TranscriptTranslatorMarianProcessor,
+    )
+
+    mock_result = {"text": {"en": "Hello world", "fr": "Bonjour le monde"}}
+
+    processor = TranscriptTranslatorMarianProcessor()
+
+    def fake_get_pref(key, default=None):
+        prefs = {"audio:source_language": "en", "audio:target_language": "fr"}
+        return prefs.get(key, default)
+
+    with (
+        patch.object(processor, "get_pref", side_effect=fake_get_pref),
+        patch(
+            "reflector.processors.transcript_translator_marian.translator_service"
+        ) as mock_svc,
+    ):
+        mock_svc.translate.return_value = mock_result
+
+        result = await processor._translate("Hello world")
+
+        assert result == "Bonjour le monde"
+        mock_svc.translate.assert_called_once_with("Hello world", "en", "fr")
+
+
+@pytest.mark.asyncio
+async def test_transcript_translator_marian_no_translation():
+    """Verify translator returns None when target language not in result."""
+    from reflector.processors.transcript_translator_marian import (
+        TranscriptTranslatorMarianProcessor,
+    )
+
+    mock_result = {"text": {"en": "Hello world"}}
+
+    processor = TranscriptTranslatorMarianProcessor()
+
+    def fake_get_pref(key, default=None):
+        prefs = {"audio:source_language": "en", "audio:target_language": "fr"}
+        return prefs.get(key, default)
+
+    with (
+        patch.object(processor, "get_pref", side_effect=fake_get_pref),
+        patch(
+            "reflector.processors.transcript_translator_marian.translator_service"
+        ) as mock_svc,
+    ):
+        mock_svc.translate.return_value = mock_result
+
+        result = await processor._translate("Hello world")
+
+        assert result is None
+
+
+# ── File Transcript Whisper Processor Tests ─────────────────────────────
+
+
+@pytest.mark.asyncio
+async def test_file_transcript_whisper_transcript():
+    """Verify whisper file processor downloads, transcribes, and returns Transcript."""
+    from reflector.processors.file_transcript import FileTranscriptInput
+    from reflector.processors.file_transcript_whisper import (
+        FileTranscriptWhisperProcessor,
+    )
+
+    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
+    tmp.write(b"fake audio")
+    tmp.close()
+    tmp_path = Path(tmp.name)
+
+    processor = FileTranscriptWhisperProcessor()
+
+    # Mock the blocking transcription method
+    mock_transcript = Transcript(
+        words=[
+            Word(text="Hello", start=0.0, end=0.5),
+            Word(text=" world", start=0.5, end=1.0),
+        ]
+    )
+
+    with (
+        patch(
+            "reflector.processors.file_transcript_whisper.download_audio_to_temp",
+            new_callable=AsyncMock,
+            return_value=tmp_path,
+        ),
+        patch.object(
+            processor,
+            "_transcribe_file_blocking",
+            return_value=mock_transcript,
+        ),
+    ):
+        data = FileTranscriptInput(
+            audio_url="https://example.com/test.wav", language="en"
+        )
+        result = await processor._transcript(data)
+
+        assert isinstance(result, Transcript)
+        assert len(result.words) == 2
+        assert result.words[0].text == "Hello"
+
+
+# ── VAD Helper Tests ────────────────────────────────────────────────────
+
+
+def test_enforce_word_timing_constraints():
+    """Verify word timing enforcement prevents overlapping times."""
+    from reflector.processors.file_transcript_whisper import (
+        _enforce_word_timing_constraints,
+    )
+
+    words = [
+        {"word": "hello", "start": 0.0, "end": 1.5},
+        {"word": "world", "start": 1.0, "end": 2.0},  # overlaps with previous
+        {"word": "test", "start": 2.0, "end": 3.0},
+    ]
+
+    result = _enforce_word_timing_constraints(words)
+
+    assert result[0]["end"] == 1.0  # Clamped to next word's start
+    assert result[1]["end"] == 2.0  # Clamped to next word's start
+    assert result[2]["end"] == 3.0  # Last word unchanged
+
+
+def test_enforce_word_timing_constraints_empty():
+    """Verify timing enforcement handles empty and single-word lists."""
+    from reflector.processors.file_transcript_whisper import (
+        _enforce_word_timing_constraints,
+    )
+
+    assert _enforce_word_timing_constraints([]) == []
+    assert _enforce_word_timing_constraints([{"word": "a", "start": 0, "end": 1}]) == [
+        {"word": "a", "start": 0, "end": 1}
+    ]
+
+
+def test_pad_audio_short():
+    """Verify short audio gets padded with silence."""
+    import numpy as np
+
+    from reflector.processors.file_transcript_whisper import _pad_audio
+
+    short_audio = np.zeros(100, dtype=np.float32)  # Very short
+    result = _pad_audio(short_audio, sample_rate=16000)
+
+    # Should be padded to at least silence_padding duration
+    assert len(result) > len(short_audio)
+
+
+def test_pad_audio_long():
+    """Verify long audio is not padded."""
+    import numpy as np
+
+    from reflector.processors.file_transcript_whisper import _pad_audio
+
+    long_audio = np.zeros(32000, dtype=np.float32)  # 2 seconds
+    result = _pad_audio(long_audio, sample_rate=16000)
+
+    assert len(result) == len(long_audio)
+
+
+# ── Translator Service Tests ────────────────────────────────────────────
+
+
+def test_translator_service_resolve_model():
+    """Verify model resolution for known and unknown language pairs."""
+    from reflector.processors._marian_translator_service import MarianTranslatorService
+
+    svc = MarianTranslatorService()
+
+    assert svc._resolve_model_name("en", "fr") == "Helsinki-NLP/opus-mt-en-fr"
+    assert svc._resolve_model_name("es", "en") == "Helsinki-NLP/opus-mt-es-en"
+    assert svc._resolve_model_name("en", "de") == "Helsinki-NLP/opus-mt-en-de"
+    # Unknown pair falls back to en->fr
+    assert svc._resolve_model_name("ja", "ko") == "Helsinki-NLP/opus-mt-en-fr"
+
+
+# ── Diarization Service Tests ───────────────────────────────────────────
+
+
+def test_diarization_service_singleton():
+    """Verify diarization_service is a module-level singleton."""
+    from reflector.processors._pyannote_diarization_service import (
+        PyannoteDiarizationService,
+        diarization_service,
+    )
+
+    assert isinstance(diarization_service, PyannoteDiarizationService)
+    assert diarization_service._pipeline is None  # Not loaded until first use
+
+
+def test_translator_service_singleton():
+    """Verify translator_service is a module-level singleton."""
+    from reflector.processors._marian_translator_service import (
+        MarianTranslatorService,
+        translator_service,
+    )
+
+    assert isinstance(translator_service, MarianTranslatorService)
+    assert translator_service._pipeline is None  # Not loaded until first use

server/tests/test_retry_decorator.py

@@ -49,6 +49,15 @@ async def test_retry_httpx(httpx_mock):
             )
 
 
+@pytest.mark.asyncio
+async def test_retry_402_stops_by_default(httpx_mock):
+    """402 (payment required / no credits) is in default retry_httpx_status_stop — do not retry."""
+    httpx_mock.add_response(status_code=402, json={"error": "insufficient_credits"})
+    async with httpx.AsyncClient() as client:
+        with pytest.raises(RetryHTTPException):
+            await retry(client.get)("https://test_url", retry_timeout=5)
+
+
 @pytest.mark.asyncio
 async def test_retry_normal():
     left = 3

server/tests/test_security_permissions.py

@@ -5,6 +5,7 @@ import time
 from pathlib import Path
 
 import pytest
+from conftest import authenticated_client_ctx
 from httpx_ws import aconnect_ws
 from uvicorn import Config, Server
 
@@ -382,3 +383,607 @@ async def test_audio_mp3_requires_token_for_owned_transcript(
     )
     resp2 = await client.get(f"/transcripts/{t.id}/audio/mp3", params={"token": token})
     assert resp2.status_code == 200
+
+
+# ======================================================================
+# Auth guards: anonymous blocked when PUBLIC_MODE=False
+# ======================================================================
+
+
+@pytest.mark.asyncio
+async def test_anonymous_cannot_create_transcript_when_not_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    resp = await client.post("/transcripts", json={"name": "anon-test"})
+    assert resp.status_code == 401, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_cannot_process_transcript_when_not_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="process-test",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="public",
+    )
+
+    resp = await client.post(f"/transcripts/{t.id}/process")
+    assert resp.status_code == 401, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_cannot_upload_when_not_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="upload-test",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="public",
+    )
+
+    # Minimal multipart upload
+    resp = await client.post(
+        f"/transcripts/{t.id}/record/upload",
+        params={"chunk_number": 0, "total_chunks": 1},
+        files={"chunk": ("test.mp3", b"fake-audio", "audio/mpeg")},
+    )
+    assert resp.status_code == 401, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_cannot_webrtc_record_when_not_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="webrtc-test",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="public",
+    )
+
+    resp = await client.post(
+        f"/transcripts/{t.id}/record/webrtc",
+        json={"sdp": "v=0\r\n", "type": "offer"},
+    )
+    assert resp.status_code == 401, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_cannot_start_meeting_recording_when_not_public(
+    client, monkeypatch
+):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    room = await rooms_controller.add(
+        name="recording-auth-test",
+        user_id="owner-rec",
+        zulip_auto_post=False,
+        zulip_stream="",
+        zulip_topic="",
+        is_locked=False,
+        room_mode="normal",
+        recording_type="cloud",
+        recording_trigger="automatic-2nd-participant",
+        is_shared=True,
+        webhook_url="",
+        webhook_secret="",
+    )
+
+    meeting = await meetings_controller.create(
+        id="meeting-rec-test",
+        room_name="recording-auth-test",
+        room_url="room-url",
+        host_room_url="host-url",
+        start_date=Room.model_fields["created_at"].default_factory(),
+        end_date=Room.model_fields["created_at"].default_factory(),
+        room=room,
+    )
+
+    resp = await client.post(
+        f"/meetings/{meeting.id}/recordings/start",
+        json={"type": "cloud", "instanceId": "00000000-0000-0000-0000-000000000001"},
+    )
+    assert resp.status_code == 401, resp.text
+
+
+# ======================================================================
+# Public mode: anonymous IS allowed when PUBLIC_MODE=True
+# ======================================================================
+
+
+@pytest.mark.asyncio
+async def test_anonymous_can_create_transcript_when_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    resp = await client.post("/transcripts", json={"name": "anon-public-test"})
+    assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_can_list_transcripts_when_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    resp = await client.get("/transcripts")
+    assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_can_read_public_transcript(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    t = await transcripts_controller.add(
+        name="readable-test",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="public",
+    )
+
+    resp = await client.get(f"/transcripts/{t.id}")
+    assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_can_upload_when_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    t = await transcripts_controller.add(
+        name="upload-public-test",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="public",
+    )
+
+    resp = await client.post(
+        f"/transcripts/{t.id}/record/upload",
+        params={"chunk_number": 0, "total_chunks": 2},
+        files={"chunk": ("test.mp3", b"fake-audio", "audio/mpeg")},
+    )
+    # Chunk 0 of 2 won't trigger av.open validation, so should succeed with "ok"
+    # The key assertion: auth did NOT block us (no 401)
+    assert resp.status_code != 401, f"Should not get 401 in public mode: {resp.text}"
+    assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_can_start_meeting_recording_when_public(client, monkeypatch):
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    room = await rooms_controller.add(
+        name="recording-public-test",
+        user_id="owner-pub",
+        zulip_auto_post=False,
+        zulip_stream="",
+        zulip_topic="",
+        is_locked=False,
+        room_mode="normal",
+        recording_type="cloud",
+        recording_trigger="automatic-2nd-participant",
+        is_shared=True,
+        webhook_url="",
+        webhook_secret="",
+    )
+
+    meeting = await meetings_controller.create(
+        id="meeting-pub-test",
+        room_name="recording-public-test",
+        room_url="room-url",
+        host_room_url="host-url",
+        start_date=Room.model_fields["created_at"].default_factory(),
+        end_date=Room.model_fields["created_at"].default_factory(),
+        room=room,
+    )
+
+    resp = await client.post(
+        f"/meetings/{meeting.id}/recordings/start",
+        json={"type": "cloud", "instanceId": "00000000-0000-0000-0000-000000000002"},
+    )
+    # Should not be 401 (may fail for other reasons like no Daily API, but auth passes)
+    assert resp.status_code != 401, f"Should not get 401 in public mode: {resp.text}"
+
+
+# ======================================================================
+# Authenticated user vs private data (own transcripts)
+# Authenticated owner should be able to create, read, and process
+# their own private transcripts even when PUBLIC_MODE=False
+# ======================================================================
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_create_transcript_private_mode(client, monkeypatch):
+    """Authenticated user can create transcripts even when PUBLIC_MODE=False."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    async with authenticated_client_ctx():
+        resp = await client.post("/transcripts", json={"name": "auth-private-create"})
+        assert resp.status_code == 200, resp.text
+        assert resp.json()["user_id"] == "randomuserid"
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_read_own_private_transcript(client, monkeypatch):
+    """Authenticated owner can read their own private transcript."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    # Create transcript owned by "randomuserid"
+    t = await transcripts_controller.add(
+        name="auth-private-read",
+        source_kind=SourceKind.LIVE,
+        user_id="randomuserid",
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.get(f"/transcripts/{t.id}")
+        assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_authenticated_cannot_read_others_private_transcript(client, monkeypatch):
+    """Authenticated user cannot read another user's private transcript."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    # Create transcript owned by someone else
+    t = await transcripts_controller.add(
+        name="other-private",
+        source_kind=SourceKind.LIVE,
+        user_id="other-owner",
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.get(f"/transcripts/{t.id}")
+        assert resp.status_code == 403, resp.text
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_process_own_private_transcript(client, monkeypatch):
+    """Authenticated owner can trigger processing on their own private transcript."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="auth-private-process",
+        source_kind=SourceKind.LIVE,
+        user_id="randomuserid",
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.post(f"/transcripts/{t.id}/process")
+        # Should pass auth (may fail for other reasons like validation, but not 401/403)
+        assert resp.status_code not in (401, 403), resp.text
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_upload_to_own_private_transcript(client, monkeypatch):
+    """Authenticated owner can upload audio to their own private transcript."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="auth-private-upload",
+        source_kind=SourceKind.LIVE,
+        user_id="randomuserid",
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.post(
+            f"/transcripts/{t.id}/record/upload",
+            params={"chunk_number": 0, "total_chunks": 2},
+            files={"chunk": ("test.mp3", b"fake-audio", "audio/mpeg")},
+        )
+        # Auth passes, chunk accepted (not final chunk so no av validation)
+        assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_webrtc_own_private_transcript(client, monkeypatch):
+    """Authenticated owner can start WebRTC recording on their own private transcript."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="auth-private-webrtc",
+        source_kind=SourceKind.LIVE,
+        user_id="randomuserid",
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.post(
+            f"/transcripts/{t.id}/record/webrtc",
+            json={"sdp": "v=0\r\n", "type": "offer"},
+        )
+        # Auth passes (may fail for other reasons like RTC setup, but not 401/403)
+        assert resp.status_code not in (401, 403), resp.text
+
+
+# ======================================================================
+# Authenticated user vs semi-private data (other user's transcripts)
+# Any authenticated user should be able to READ semi-private transcripts
+# but NOT write to them (upload, process) since they don't own them
+# ======================================================================
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_read_others_semi_private_transcript(
+    client, monkeypatch
+):
+    """Any authenticated user can read a semi-private transcript (link sharing)."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    # Create transcript owned by someone else with semi-private share mode
+    t = await transcripts_controller.add(
+        name="semi-private-readable",
+        source_kind=SourceKind.LIVE,
+        user_id="other-owner",
+        share_mode="semi-private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.get(f"/transcripts/{t.id}")
+        assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_cannot_read_semi_private_transcript(client, monkeypatch):
+    """Anonymous user cannot read a semi-private transcript."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="semi-private-blocked",
+        source_kind=SourceKind.LIVE,
+        user_id="some-owner",
+        share_mode="semi-private",
+    )
+
+    resp = await client.get(f"/transcripts/{t.id}")
+    assert resp.status_code == 403, resp.text
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_list_own_transcripts_private_mode(client, monkeypatch):
+    """Authenticated user can list their own transcripts when PUBLIC_MODE=False."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    await transcripts_controller.add(
+        name="my-transcript",
+        source_kind=SourceKind.LIVE,
+        user_id="randomuserid",
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.get("/transcripts")
+        assert resp.status_code == 200, resp.text
+        items = resp.json()["items"]
+        assert len(items) >= 1
+        # All returned transcripts should belong to the user or be in shared rooms
+        for item in items:
+            assert item["user_id"] == "randomuserid" or item.get("room_id") is not None
+
+
+@pytest.mark.asyncio
+async def test_authenticated_cannot_list_others_private_transcripts(
+    client, monkeypatch
+):
+    """Authenticated user should NOT see another user's private transcripts in the list."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    await transcripts_controller.add(
+        name="hidden-from-others",
+        source_kind=SourceKind.LIVE,
+        user_id="secret-owner",
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.get("/transcripts")
+        assert resp.status_code == 200, resp.text
+        items = resp.json()["items"]
+        # Should not contain transcripts owned by "secret-owner"
+        for item in items:
+            assert (
+                item.get("user_id") != "secret-owner"
+            ), f"Leaked private transcript: {item['id']}"
+
+
+# ======================================================================
+# Anonymous-created transcripts (user_id=None)
+# These transcripts bypass share_mode checks entirely in get_by_id_for_http.
+# They should always be accessible to everyone regardless of PUBLIC_MODE
+# or share_mode setting, because there is no owner to restrict access.
+# ======================================================================
+
+
+@pytest.mark.asyncio
+async def test_anonymous_transcript_accessible_when_public_mode_true(
+    client, monkeypatch
+):
+    """Anonymous transcript (user_id=None) is accessible even with default private share_mode
+    when PUBLIC_MODE=True."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    t = await transcripts_controller.add(
+        name="anon-transcript-public-mode",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="private",  # share_mode is irrelevant for user_id=None
+    )
+
+    resp = await client.get(f"/transcripts/{t.id}")
+    assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_transcript_accessible_when_public_mode_false(
+    client, monkeypatch
+):
+    """Anonymous transcript (user_id=None) is accessible by authenticated users
+    even when PUBLIC_MODE=False. The transcript has no owner, so share_mode is bypassed."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="anon-transcript-private-mode",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.get(f"/transcripts/{t.id}")
+        assert resp.status_code == 200, resp.text
+
+
+@pytest.mark.asyncio
+async def test_anonymous_transcript_accessible_regardless_of_share_mode(
+    client, monkeypatch
+):
+    """Anonymous transcripts (user_id=None) are accessible regardless of share_mode value.
+    Tests all three share modes to confirm the user_id=None bypass works consistently."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    for mode in ("private", "semi-private", "public"):
+        t = await transcripts_controller.add(
+            name=f"anon-share-{mode}",
+            source_kind=SourceKind.LIVE,
+            user_id=None,
+            share_mode=mode,
+        )
+
+        resp = await client.get(f"/transcripts/{t.id}")
+        assert resp.status_code == 200, f"Failed for share_mode={mode}: {resp.text}"
+
+
+@pytest.mark.asyncio
+async def test_anonymous_transcript_readable_by_different_authenticated_user(
+    client, monkeypatch
+):
+    """An authenticated user can read anonymous transcripts (user_id=None) even with
+    private share_mode, because the no-owner bypass applies."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="anon-read-by-auth-user",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="private",
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.get(f"/transcripts/{t.id}")
+        assert resp.status_code == 200, resp.text
+        assert resp.json()["user_id"] is None
+
+
+@pytest.mark.asyncio
+async def test_anonymous_transcript_in_list_when_public_mode(client, monkeypatch):
+    """Anonymous transcripts appear in the transcript list when PUBLIC_MODE=True."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
+    t = await transcripts_controller.add(
+        name="anon-in-list",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="private",
+    )
+
+    resp = await client.get("/transcripts")
+    assert resp.status_code == 200, resp.text
+    ids = [item["id"] for item in resp.json()["items"]]
+    assert t.id in ids, "Anonymous transcript should appear in the public list"
+
+
+@pytest.mark.asyncio
+async def test_anonymous_transcript_audio_accessible(client, monkeypatch, tmpdir):
+    """Anonymous transcript audio (mp3) is accessible without authentication
+    because user_id=None bypasses share_mode checks."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+    monkeypatch.setattr(settings, "DATA_DIR", Path(tmpdir).as_posix())
+
+    t = await transcripts_controller.add(
+        name="anon-audio-access",
+        source_kind=SourceKind.LIVE,
+        user_id=None,
+        share_mode="private",
+    )
+
+    tr = await transcripts_controller.get_by_id(t.id)
+    await transcripts_controller.update(tr, {"status": "ended"})
+
+    # Copy fixture audio to transcript path
+    audio_path = Path(__file__).parent / "records" / "test_mathieu_hello.mp3"
+    tr.audio_mp3_filename.parent.mkdir(parents=True, exist_ok=True)
+    shutil.copy(audio_path, tr.audio_mp3_filename)
+
+    resp = await client.get(f"/transcripts/{t.id}/audio/mp3")
+    assert (
+        resp.status_code == 200
+    ), f"Anonymous transcript audio should be accessible: {resp.text}"
+
+
+@pytest.mark.asyncio
+async def test_owned_transcript_not_accessible_by_anon_when_not_public(
+    client, monkeypatch
+):
+    """Contrast test: owned transcript with private share_mode is NOT accessible
+    to anonymous users when PUBLIC_MODE=False. This confirms that the user_id=None
+    bypass only applies to anonymous transcripts, not to all transcripts."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    t = await transcripts_controller.add(
+        name="owned-private-contrast",
+        source_kind=SourceKind.LIVE,
+        user_id="some-owner",
+        share_mode="private",
+    )
+
+    resp = await client.get(f"/transcripts/{t.id}")
+    assert (
+        resp.status_code == 403
+    ), f"Owned private transcript should be denied to anonymous: {resp.text}"
+
+
+@pytest.mark.asyncio
+async def test_authenticated_can_start_meeting_recording_private_mode(
+    client, monkeypatch
+):
+    """Authenticated user can start recording in non-public mode."""
+    monkeypatch.setattr(settings, "PUBLIC_MODE", False)
+
+    room = await rooms_controller.add(
+        name="auth-recording-test",
+        user_id="randomuserid",
+        zulip_auto_post=False,
+        zulip_stream="",
+        zulip_topic="",
+        is_locked=False,
+        room_mode="normal",
+        recording_type="cloud",
+        recording_trigger="automatic-2nd-participant",
+        is_shared=True,
+        webhook_url="",
+        webhook_secret="",
+    )
+
+    meeting = await meetings_controller.create(
+        id="meeting-auth-rec",
+        room_name="auth-recording-test",
+        room_url="room-url",
+        host_room_url="host-url",
+        start_date=Room.model_fields["created_at"].default_factory(),
+        end_date=Room.model_fields["created_at"].default_factory(),
+        room=room,
+    )
+
+    async with authenticated_client_ctx():
+        resp = await client.post(
+            f"/meetings/{meeting.id}/recordings/start",
+            json={
+                "type": "cloud",
+                "instanceId": "00000000-0000-0000-0000-000000000003",
+            },
+        )
+        # Auth passes (may fail for Daily API reasons, but not 401)
+        assert resp.status_code != 401, resp.text

server/tests/test_transcript_formats.py

@@ -340,8 +340,13 @@ async def test_transcript_formats_with_overlapping_speakers_multitrack():
 
 
 @pytest.mark.asyncio
-async def test_api_transcript_format_text(client):
+async def test_api_transcript_format_text(monkeypatch, client):
     """Test GET /transcripts/{id} with transcript_format=text."""
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post("/transcripts", json={"name": "Test transcript"})
     assert response.status_code == 200
     tid = response.json()["id"]
@@ -390,8 +395,13 @@ async def test_api_transcript_format_text(client):
 
 
 @pytest.mark.asyncio
-async def test_api_transcript_format_text_timestamped(client):
+async def test_api_transcript_format_text_timestamped(monkeypatch, client):
     """Test GET /transcripts/{id} with transcript_format=text-timestamped."""
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post("/transcripts", json={"name": "Test transcript"})
     assert response.status_code == 200
     tid = response.json()["id"]
@@ -441,8 +451,13 @@ async def test_api_transcript_format_text_timestamped(client):
 
 
 @pytest.mark.asyncio
-async def test_api_transcript_format_webvtt_named(client):
+async def test_api_transcript_format_webvtt_named(monkeypatch, client):
     """Test GET /transcripts/{id} with transcript_format=webvtt-named."""
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post("/transcripts", json={"name": "Test transcript"})
     assert response.status_code == 200
     tid = response.json()["id"]
@@ -491,8 +506,13 @@ async def test_api_transcript_format_webvtt_named(client):
 
 
 @pytest.mark.asyncio
-async def test_api_transcript_format_json(client):
+async def test_api_transcript_format_json(monkeypatch, client):
     """Test GET /transcripts/{id} with transcript_format=json."""
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post("/transcripts", json={"name": "Test transcript"})
     assert response.status_code == 200
     tid = response.json()["id"]
@@ -544,8 +564,13 @@ async def test_api_transcript_format_json(client):
 
 
 @pytest.mark.asyncio
-async def test_api_transcript_format_default_is_text(client):
+async def test_api_transcript_format_default_is_text(monkeypatch, client):
     """Test GET /transcripts/{id} defaults to text format."""
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post("/transcripts", json={"name": "Test transcript"})
     assert response.status_code == 200
     tid = response.json()["id"]
@@ -654,12 +679,18 @@ async def test_api_topics_endpoint_multitrack_segmentation(client):
 
 
 @pytest.mark.asyncio
-async def test_api_topics_endpoint_non_multitrack_segmentation(client):
+async def test_api_topics_endpoint_non_multitrack_segmentation(monkeypatch, client):
     """Test GET /transcripts/{id}/topics uses default segmentation for non-multitrack.
 
     Ensures backward compatibility - transcripts without multitrack recordings
     should continue using the default speaker-change-based segmentation.
     """
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
+
     from reflector.db.transcripts import (
         TranscriptParticipant,
         TranscriptTopic,

server/tests/test_transcripts.py

@@ -5,7 +5,12 @@ from reflector.db.transcripts import transcripts_controller
 
 
 @pytest.mark.asyncio
-async def test_transcript_create(client):
+async def test_transcript_create(monkeypatch, client):
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
     assert response.json()["name"] == "test"
@@ -110,6 +115,33 @@ async def test_transcript_get_update_title(authenticated_client, client):
     assert response.json()["title"] == "test_title"
 
 
+@pytest.mark.asyncio
+async def test_set_status_emits_status_event_and_updates_transcript(
+    monkeypatch, client
+):
+    """set_status adds a STATUS event and updates the transcript status (broadcast for WebSocket)."""
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
+    response = await client.post("/transcripts", json={"name": "Status test"})
+    assert response.status_code == 200
+    transcript_id = response.json()["id"]
+
+    transcript = await transcripts_controller.get_by_id(transcript_id)
+    assert transcript is not None
+    assert transcript.status == "idle"
+
+    event = await transcripts_controller.set_status(transcript_id, "processing")
+    assert event is not None
+    assert event.event == "STATUS"
+    assert event.data.get("value") == "processing"
+
+    updated = await transcripts_controller.get_by_id(transcript_id)
+    assert updated.status == "processing"
+
+
 @pytest.mark.asyncio
 async def test_transcripts_list_anonymous(client):
     # XXX this test is a bit fragile, as it depends on the storage which
@@ -233,3 +265,43 @@ async def test_transcript_get_returns_null_room_name_when_no_room(
     assert response.status_code == 200
     assert response.json()["room_id"] is None
     assert response.json()["room_name"] is None
+
+
+@pytest.mark.asyncio
+async def test_transcripts_list_filtered_by_room_id(authenticated_client, client):
+    """GET /transcripts?room_id=X returns only transcripts for that room."""
+    # Use same user as authenticated_client (conftest uses "randomuserid")
+    user_id = "randomuserid"
+    room = await rooms_controller.add(
+        name="room-for-list-filter",
+        user_id=user_id,
+        zulip_auto_post=False,
+        zulip_stream="",
+        zulip_topic="",
+        is_locked=False,
+        room_mode="normal",
+        recording_type="cloud",
+        recording_trigger="automatic-2nd-participant",
+        is_shared=False,
+        webhook_url="",
+        webhook_secret="",
+    )
+    in_room = await transcripts_controller.add(
+        name="in-room",
+        source_kind="file",
+        room_id=room.id,
+        user_id=user_id,
+    )
+    other = await transcripts_controller.add(
+        name="no-room",
+        source_kind="file",
+        room_id=None,
+        user_id=user_id,
+    )
+
+    response = await client.get("/transcripts", params={"room_id": room.id})
+    assert response.status_code == 200
+    items = response.json()["items"]
+    ids = [t["id"] for t in items]
+    assert in_room.id in ids
+    assert other.id not in ids

server/tests/test_transcripts_audio_download.py

@@ -5,10 +5,13 @@ import pytest
 
 
 @pytest.fixture
-async def fake_transcript(tmpdir, client):
+async def fake_transcript(tmpdir, client, monkeypatch):
     from reflector.settings import settings
     from reflector.views.transcripts import transcripts_controller
 
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     settings.DATA_DIR = Path(tmpdir)
 
     # create a transcript

server/tests/test_transcripts_audio_token_auth.py

@@ -0,0 +1,327 @@
+"""Tests for audio mp3 endpoint token query-param authentication.
+
+Covers both password (HS256) and JWT/Authentik (RS256) auth backends,
+verifying that private transcripts can be accessed via ?token= query param.
+"""
+
+import shutil
+from datetime import datetime, timedelta, timezone
+from pathlib import Path
+from unittest.mock import patch
+
+import jwt
+import pytest
+from cryptography.hazmat.primitives import serialization
+from cryptography.hazmat.primitives.asymmetric import rsa
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+OWNER_USER_ID = "test-owner-user-id"
+
+
+def _create_hs256_token(user_id: str, secret: str, expired: bool = False) -> str:
+    """Create an HS256 JWT like the password auth backend does."""
+    delta = timedelta(minutes=-5) if expired else timedelta(hours=24)
+    payload = {
+        "sub": user_id,
+        "email": "test@example.com",
+        "exp": datetime.now(timezone.utc) + delta,
+    }
+    return jwt.encode(payload, secret, algorithm="HS256")
+
+
+def _generate_rsa_keypair():
+    """Generate a fresh RSA keypair for tests."""
+    private_key = rsa.generate_private_key(public_exponent=65537, key_size=2048)
+    public_pem = private_key.public_key().public_bytes(
+        serialization.Encoding.PEM, serialization.PublicFormat.SubjectPublicKeyInfo
+    )
+    return private_key, public_pem.decode()
+
+
+def _create_rs256_token(
+    authentik_uid: str,
+    private_key,
+    audience: str,
+    expired: bool = False,
+) -> str:
+    """Create an RS256 JWT like Authentik would issue."""
+    delta = timedelta(minutes=-5) if expired else timedelta(hours=1)
+    payload = {
+        "sub": authentik_uid,
+        "email": "authentik-user@example.com",
+        "aud": audience,
+        "exp": datetime.now(timezone.utc) + delta,
+    }
+    return jwt.encode(payload, private_key, algorithm="RS256")
+
+
+# ---------------------------------------------------------------------------
+# Fixtures
+# ---------------------------------------------------------------------------
+
+
+@pytest.fixture
+async def private_transcript(tmpdir):
+    """Create a private transcript owned by OWNER_USER_ID with an mp3 file.
+
+    Created directly via the controller (not HTTP) so no auth override
+    leaks into the test scope.
+    """
+    from reflector.db.transcripts import SourceKind, transcripts_controller
+    from reflector.settings import settings
+
+    settings.DATA_DIR = Path(tmpdir)
+
+    transcript = await transcripts_controller.add(
+        "Private audio test",
+        source_kind=SourceKind.FILE,
+        user_id=OWNER_USER_ID,
+        share_mode="private",
+    )
+    await transcripts_controller.update(transcript, {"status": "ended"})
+
+    # Copy a real mp3 to the expected location
+    audio_filename = transcript.audio_mp3_filename
+    mp3_source = Path(__file__).parent / "records" / "test_mathieu_hello.mp3"
+    audio_filename.parent.mkdir(parents=True, exist_ok=True)
+    shutil.copy(mp3_source, audio_filename)
+
+    yield transcript
+
+
+# ---------------------------------------------------------------------------
+# Core access control tests
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_private_no_auth_returns_403(private_transcript, client):
+    """Without auth, accessing a private transcript's audio returns 403."""
+    response = await client.get(f"/transcripts/{private_transcript.id}/audio/mp3")
+    assert response.status_code == 403
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_with_bearer_header(private_transcript, client):
+    """Owner accessing audio via Authorization header works."""
+    from reflector.app import app
+    from reflector.auth import current_user_optional
+
+    # Temporarily override to simulate the owner being authenticated
+    app.dependency_overrides[current_user_optional] = lambda: {
+        "sub": OWNER_USER_ID,
+        "email": "test@example.com",
+    }
+    try:
+        response = await client.get(f"/transcripts/{private_transcript.id}/audio/mp3")
+    finally:
+        del app.dependency_overrides[current_user_optional]
+
+    assert response.status_code == 200
+    assert response.headers["content-type"] == "audio/mpeg"
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_public_transcript_no_auth_ok(tmpdir, client):
+    """Public transcripts are accessible without any auth."""
+    from reflector.db.transcripts import SourceKind, transcripts_controller
+    from reflector.settings import settings
+
+    settings.DATA_DIR = Path(tmpdir)
+
+    transcript = await transcripts_controller.add(
+        "Public audio test",
+        source_kind=SourceKind.FILE,
+        user_id=OWNER_USER_ID,
+        share_mode="public",
+    )
+    await transcripts_controller.update(transcript, {"status": "ended"})
+
+    audio_filename = transcript.audio_mp3_filename
+    mp3_source = Path(__file__).parent / "records" / "test_mathieu_hello.mp3"
+    audio_filename.parent.mkdir(parents=True, exist_ok=True)
+    shutil.copy(mp3_source, audio_filename)
+
+    response = await client.get(f"/transcripts/{transcript.id}/audio/mp3")
+    assert response.status_code == 200
+    assert response.headers["content-type"] == "audio/mpeg"
+
+
+# ---------------------------------------------------------------------------
+# Password auth backend tests (?token= with HS256)
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_password_token_query_param(private_transcript, client):
+    """Password backend: valid HS256 ?token= grants access to private audio."""
+    from reflector.auth.auth_password import UserInfo
+    from reflector.settings import settings
+
+    token = _create_hs256_token(OWNER_USER_ID, settings.SECRET_KEY)
+
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        mock_verify.return_value = UserInfo(sub=OWNER_USER_ID, email="test@example.com")
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3?token={token}"
+        )
+
+    assert response.status_code == 200
+    assert response.headers["content-type"] == "audio/mpeg"
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_password_expired_token_returns_401(private_transcript, client):
+    """Password backend: expired HS256 ?token= returns 401."""
+    from reflector.settings import settings
+
+    expired_token = _create_hs256_token(
+        OWNER_USER_ID, settings.SECRET_KEY, expired=True
+    )
+
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        mock_verify.side_effect = jwt.ExpiredSignatureError("token expired")
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3" f"?token={expired_token}"
+        )
+
+    assert response.status_code == 401
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_password_wrong_user_returns_403(private_transcript, client):
+    """Password backend: valid token for a different user returns 403."""
+    from reflector.auth.auth_password import UserInfo
+    from reflector.settings import settings
+
+    token = _create_hs256_token("other-user-id", settings.SECRET_KEY)
+
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        mock_verify.return_value = UserInfo(
+            sub="other-user-id", email="other@example.com"
+        )
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3?token={token}"
+        )
+
+    assert response.status_code == 403
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_invalid_token_returns_401(private_transcript, client):
+    """Garbage token string returns 401."""
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        mock_verify.return_value = None
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3" "?token=not-a-real-token"
+        )
+
+    assert response.status_code == 401
+
+
+# ---------------------------------------------------------------------------
+# JWT/Authentik auth backend tests (?token= with RS256)
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_authentik_token_query_param(private_transcript, client):
+    """Authentik backend: valid RS256 ?token= grants access to private audio."""
+    from reflector.auth.auth_password import UserInfo
+
+    private_key, _ = _generate_rsa_keypair()
+    token = _create_rs256_token("authentik-abc123", private_key, "test-audience")
+
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        # Authentik flow maps authentik_uid -> internal user id
+        mock_verify.return_value = UserInfo(
+            sub=OWNER_USER_ID, email="authentik-user@example.com"
+        )
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3?token={token}"
+        )
+
+    assert response.status_code == 200
+    assert response.headers["content-type"] == "audio/mpeg"
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_authentik_expired_token_returns_401(
+    private_transcript, client
+):
+    """Authentik backend: expired RS256 ?token= returns 401."""
+    private_key, _ = _generate_rsa_keypair()
+    expired_token = _create_rs256_token(
+        "authentik-abc123", private_key, "test-audience", expired=True
+    )
+
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        mock_verify.side_effect = jwt.ExpiredSignatureError("token expired")
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3" f"?token={expired_token}"
+        )
+
+    assert response.status_code == 401
+
+
+@pytest.mark.asyncio
+async def test_audio_mp3_authentik_wrong_user_returns_403(private_transcript, client):
+    """Authentik backend: valid RS256 token for different user returns 403."""
+    from reflector.auth.auth_password import UserInfo
+
+    private_key, _ = _generate_rsa_keypair()
+    token = _create_rs256_token("authentik-other", private_key, "test-audience")
+
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        mock_verify.return_value = UserInfo(
+            sub="different-user-id", email="other@example.com"
+        )
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3?token={token}"
+        )
+
+    assert response.status_code == 403
+
+
+# ---------------------------------------------------------------------------
+# _generate_local_audio_link produces HS256 tokens — must be verifiable
+# by any auth backend
+# ---------------------------------------------------------------------------
+
+
+@pytest.mark.asyncio
+async def test_local_audio_link_token_works_with_authentik_backend(
+    private_transcript, client
+):
+    """_generate_local_audio_link creates an HS256 token via create_access_token.
+
+    When the Authentik (RS256) auth backend is active, verify_raw_token uses
+    JWTAuth which expects RS256 + public key. The HS256 token created by
+    _generate_local_audio_link will fail verification, returning 401.
+
+    This test documents the bug: the internal audio URL generated for the
+    diarization pipeline is unusable under the JWT auth backend.
+    """
+    from urllib.parse import parse_qs, urlparse
+
+    # Generate the internal audio link (uses create_access_token → HS256)
+    url = private_transcript._generate_local_audio_link()
+    parsed = urlparse(url)
+    token = parse_qs(parsed.query)["token"][0]
+
+    # Simulate what happens when the JWT/Authentik backend tries to verify
+    # this HS256 token: JWTAuth.verify_token expects RS256, so it raises.
+    with patch("reflector.auth.verify_raw_token") as mock_verify:
+        mock_verify.side_effect = jwt.exceptions.InvalidAlgorithmError(
+            "the specified alg value is not allowed"
+        )
+        response = await client.get(
+            f"/transcripts/{private_transcript.id}/audio/mp3?token={token}"
+        )
+
+    # BUG: this should be 200 (the token was created by our own server),
+    # but the Authentik backend rejects it because it's HS256, not RS256.
+    assert response.status_code == 200

server/tests/test_transcripts_process.py

@@ -5,6 +5,8 @@ from unittest.mock import AsyncMock, patch
 import pytest
 from httpx import ASGITransport, AsyncClient
 
+from reflector.settings import settings
+
 
 @pytest.fixture
 async def app_lifespan():
@@ -36,7 +38,11 @@ async def test_transcript_process(
     dummy_file_diarization,
     dummy_storage,
     client,
+    monkeypatch,
 ):
+    # public mode: this test uses an anonymous client; allow anonymous transcript creation
+    monkeypatch.setattr(settings, "PUBLIC_MODE", True)
+
     # create a transcript
     response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200
@@ -106,12 +112,17 @@ async def test_transcript_process(
 
 @pytest.mark.usefixtures("setup_database")
 @pytest.mark.asyncio
-async def test_whereby_recording_uses_file_pipeline(client):
+async def test_whereby_recording_uses_file_pipeline(monkeypatch, client):
     """Test that Whereby recordings (bucket_name but no track_keys) use file pipeline"""
     from datetime import datetime, timezone
 
     from reflector.db.recordings import Recording, recordings_controller
     from reflector.db.transcripts import transcripts_controller
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
 
     # Create transcript with Whereby recording (has bucket_name, no track_keys)
     transcript = await transcripts_controller.add(
@@ -157,13 +168,18 @@ async def test_whereby_recording_uses_file_pipeline(client):
 
 @pytest.mark.usefixtures("setup_database")
 @pytest.mark.asyncio
-async def test_dailyco_recording_uses_multitrack_pipeline(client):
+async def test_dailyco_recording_uses_multitrack_pipeline(monkeypatch, client):
     """Test that Daily.co recordings (bucket_name + track_keys) use multitrack pipeline"""
     from datetime import datetime, timezone
 
     from reflector.db.recordings import Recording, recordings_controller
     from reflector.db.rooms import rooms_controller
     from reflector.db.transcripts import transcripts_controller
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
 
     room = await rooms_controller.add(
         name="test-room",
@@ -231,3 +247,86 @@ async def test_dailyco_recording_uses_multitrack_pipeline(client):
             {"s3_key": k} for k in track_keys
         ]
         mock_file_pipeline.delay.assert_not_called()
+
+
+@pytest.mark.usefixtures("setup_database")
+@pytest.mark.asyncio
+async def test_reprocess_error_transcript_passes_force(monkeypatch, client):
+    """When transcript status is 'error', reprocess passes force=True to start fresh workflow."""
+    from datetime import datetime, timezone
+
+    from reflector.db.recordings import Recording, recordings_controller
+    from reflector.db.rooms import rooms_controller
+    from reflector.db.transcripts import transcripts_controller
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
+
+    room = await rooms_controller.add(
+        name="test-room",
+        user_id="test-user",
+        zulip_auto_post=False,
+        zulip_stream="",
+        zulip_topic="",
+        is_locked=False,
+        room_mode="normal",
+        recording_type="cloud",
+        recording_trigger="automatic-2nd-participant",
+        is_shared=False,
+    )
+
+    transcript = await transcripts_controller.add(
+        "",
+        source_kind="room",
+        source_language="en",
+        target_language="en",
+        user_id="test-user",
+        share_mode="public",
+        room_id=room.id,
+    )
+
+    track_keys = ["recordings/test-room/track1.webm"]
+    recording = await recordings_controller.create(
+        Recording(
+            bucket_name="daily-bucket",
+            object_key="recordings/test-room",
+            meeting_id="test-meeting",
+            track_keys=track_keys,
+            recorded_at=datetime.now(timezone.utc),
+        )
+    )
+
+    await transcripts_controller.update(
+        transcript,
+        {
+            "recording_id": recording.id,
+            "status": "error",
+            "workflow_run_id": "old-failed-run",
+        },
+    )
+
+    with (
+        patch(
+            "reflector.services.transcript_process.task_is_scheduled_or_active"
+        ) as mock_celery,
+        patch(
+            "reflector.services.transcript_process.HatchetClientManager"
+        ) as mock_hatchet,
+        patch(
+            "reflector.views.transcripts_process.dispatch_transcript_processing",
+            new_callable=AsyncMock,
+        ) as mock_dispatch,
+    ):
+        mock_celery.return_value = False
+        from hatchet_sdk.clients.rest.models import V1TaskStatus
+
+        mock_hatchet.get_workflow_run_status = AsyncMock(
+            return_value=V1TaskStatus.FAILED
+        )
+        response = await client.post(f"/transcripts/{transcript.id}/process")
+
+    assert response.status_code == 200
+    mock_dispatch.assert_called_once()
+    assert mock_dispatch.call_args.kwargs["force"] is True

server/tests/test_transcripts_rtc_ws.py

@@ -133,10 +133,17 @@ async def test_transcript_rtc_and_websocket(
     fake_mp3_upload,
     appserver,
     client,
+    monkeypatch,
 ):
     # goal: start the server, exchange RTC, receive websocket events
     # because of that, we need to start the server in a thread
     # to be able to connect with aiortc
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
+
     server, host, port = appserver
 
     # create a transcript
@@ -298,11 +305,18 @@ async def test_transcript_rtc_and_websocket_and_fr(
     fake_mp3_upload,
     appserver,
     client,
+    monkeypatch,
 ):
     # goal: start the server, exchange RTC, receive websocket events
     # because of that, we need to start the server in a thread
     # to be able to connect with aiortc
     # with target french language
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
+
     server, host, port = appserver
 
     # create a transcript

server/tests/test_transcripts_translation.py

@@ -1,8 +1,13 @@
 import pytest
 
+from reflector.settings import settings
+
 
 @pytest.mark.asyncio
-async def test_transcript_create_default_translation(client):
+async def test_transcript_create_default_translation(monkeypatch, client):
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post("/transcripts", json={"name": "test en"})
     assert response.status_code == 200
     assert response.json()["name"] == "test en"
@@ -18,7 +23,10 @@ async def test_transcript_create_default_translation(client):
 
 
 @pytest.mark.asyncio
-async def test_transcript_create_en_fr_translation(client):
+async def test_transcript_create_en_fr_translation(monkeypatch, client):
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post(
         "/transcripts", json={"name": "test en/fr", "target_language": "fr"}
     )
@@ -36,7 +44,10 @@ async def test_transcript_create_en_fr_translation(client):
 
 
 @pytest.mark.asyncio
-async def test_transcript_create_fr_en_translation(client):
+async def test_transcript_create_fr_en_translation(monkeypatch, client):
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     response = await client.post(
         "/transcripts", json={"name": "test fr/en", "source_language": "fr"}
     )

server/tests/test_transcripts_upload.py

@@ -16,7 +16,13 @@ async def test_transcript_upload_file(
     dummy_file_diarization,
     dummy_storage,
     client,
+    monkeypatch,
 ):
+    from reflector.settings import settings
+
+    monkeypatch.setattr(
+        settings, "PUBLIC_MODE", True
+    )  # public mode: allow anonymous transcript creation for this test
     # create a transcript
     response = await client.post("/transcripts", json={"name": "test"})
     assert response.status_code == 200

server/tests/test_user_websocket_auth.py

@@ -141,33 +141,19 @@ async def test_user_ws_accepts_valid_token_and_receives_events(appserver_ws_user
         await asyncio.sleep(0.2)
 
         # Emit an event to the user's room via a standard HTTP action
+        # Use a real HTTP request to the server with the JWT token so that
+        # current_user_optional_if_public_mode is exercised without dependency overrides
         from httpx import AsyncClient
 
-        from reflector.app import app
+        async with AsyncClient(base_url=f"http://{host}:{port}/v1") as ac:
-        from reflector.auth import current_user, current_user_optional
+            resp = await ac.post(
-
+                "/transcripts",
-        # Override auth dependencies so HTTP request is performed as the same user
+                json={"name": "WS Test"},
-        # Use the internal user.id (not the Authentik UID)
+                headers={"Authorization": f"Bearer {token}"},
-        app.dependency_overrides[current_user] = lambda: {
+            )
-            "sub": user.id,
-            "email": "user-abc@example.com",
-        }
-        app.dependency_overrides[current_user_optional] = lambda: {
-            "sub": user.id,
-            "email": "user-abc@example.com",
-        }
-
-        # Use in-memory client (global singleton makes it share ws_manager)
-        async with AsyncClient(app=app, base_url=f"http://{host}:{port}/v1") as ac:
-            # Create a transcript as this user so that the server publishes TRANSCRIPT_CREATED to user room
-            resp = await ac.post("/transcripts", json={"name": "WS Test"})
             assert resp.status_code == 200
 
         # Receive the published event
         msg = await ws.receive_json()
         assert msg["event"] == "TRANSCRIPT_CREATED"
         assert "id" in msg["data"]
-
-        # Clean overrides
-        del app.dependency_overrides[current_user]
-        del app.dependency_overrides[current_user_optional]

www/app/(app)/transcripts/__tests__/webSocketReconnect.test.ts

@@ -0,0 +1,26 @@
+/**
+ * Reconnection policy for WebSocket.
+ * Ensures exponential backoff is applied and capped at 30s.
+ */
+import { getReconnectDelayMs, MAX_RETRIES } from "../webSocketReconnect";
+
+describe("webSocketReconnect", () => {
+  describe("getReconnectDelayMs", () => {
+    it("returns exponential backoff: 1s, 2s, 4s, 8s, 16s, then cap 30s", () => {
+      expect(getReconnectDelayMs(0)).toBe(1000);
+      expect(getReconnectDelayMs(1)).toBe(2000);
+      expect(getReconnectDelayMs(2)).toBe(4000);
+      expect(getReconnectDelayMs(3)).toBe(8000);
+      expect(getReconnectDelayMs(4)).toBe(16000);
+      expect(getReconnectDelayMs(5)).toBe(30000); // 32s capped to 30s
+      expect(getReconnectDelayMs(6)).toBe(30000);
+      expect(getReconnectDelayMs(9)).toBe(30000);
+    });
+
+    it("never exceeds 30s for any retry index", () => {
+      for (let i = 0; i <= MAX_RETRIES; i++) {
+        expect(getReconnectDelayMs(i)).toBeLessThanOrEqual(30000);
+      }
+    });
+  });
+});

www/app/(app)/transcripts/useMp3.ts

@@ -78,7 +78,9 @@ const useMp3 = (transcriptId: string, waiting?: boolean): Mp3Response => {
 
     // Audio is not deleted, proceed to load it
     audioElement = document.createElement("audio");
-    const audioUrl = `${API_URL}/v1/transcripts/${transcriptId}/audio/mp3`;
+    const audioUrl = accessTokenInfo
+      ? `${API_URL}/v1/transcripts/${transcriptId}/audio/mp3?token=${encodeURIComponent(accessTokenInfo)}`
+      : `${API_URL}/v1/transcripts/${transcriptId}/audio/mp3`;
     audioElement.src = audioUrl;
     audioElement.crossOrigin = "anonymous";
     audioElement.preload = "auto";

www/app/(app)/transcripts/useWebSockets.ts

@@ -14,6 +14,7 @@ import {
 } from "../../lib/apiHooks";
 import { useAuth } from "../../lib/AuthProvider";
 import { parseNonEmptyString } from "../../lib/utils";
+import { getReconnectDelayMs, MAX_RETRIES } from "./webSocketReconnect";
 
 type TranscriptWsEvent =
   operations["v1_transcript_get_websocket_events"]["responses"][200]["content"]["application/json"];
@@ -338,7 +339,6 @@ export const useWebSockets = (transcriptId: string | null): UseWebSockets => {
     if (!transcriptId) return;
     const tsId = parseNonEmptyString(transcriptId);
 
-    const MAX_RETRIES = 10;
     const url = `${WEBSOCKET_URL}/v1/transcripts/${transcriptId}/events`;
     let ws: WebSocket | null = null;
     let retryCount = 0;
@@ -472,7 +472,7 @@ export const useWebSockets = (transcriptId: string | null): UseWebSockets => {
         if (normalCodes.includes(event.code)) return;
 
         if (retryCount < MAX_RETRIES) {
-          const delay = Math.min(1000 * Math.pow(2, retryCount), 30000);
+          const delay = getReconnectDelayMs(retryCount);
           console.log(
             `WebSocket reconnecting in ${delay}ms (attempt ${retryCount + 1}/${MAX_RETRIES})`,
           );

www/app/(app)/transcripts/webSocketReconnect.ts

@@ -0,0 +1,10 @@
+/** Reconnection policy for WebSocket: exponential backoff, capped at 30s. */
+export const MAX_RETRIES = 10;
+
+/**
+ * Delay in ms before reconnecting. retryIndex is 0-based (0 = first retry).
+ * Returns 1000, 2000, 4000, ... up to 30000 max.
+ */
+export function getReconnectDelayMs(retryIndex: number): number {
+  return Math.min(1000 * Math.pow(2, retryIndex), 30000);
+}

www/package.json

@@ -10,7 +10,8 @@
     "lint": "next lint",
     "format": "prettier --write .",
     "openapi": "openapi-typescript http://127.0.0.1:1250/openapi.json -o ./app/reflector-api.d.ts",
-    "test": "jest"
+    "test": "jest",
+    "typecheck": "tsc --noEmit"
   },
   "dependencies": {
     "@chakra-ui/react": "^3.33.0",
@@ -72,7 +73,9 @@
     "overrides": {
       "minimatch@>=5.0.0 <5.1.8": "5.1.8",
       "js-yaml@<4.1.1": "4.1.1",
-      "webpack": "5.105.3"
+      "webpack": "5.105.3",
+      "serialize-javascript": "7.0.4",
+      "immutable": "5.1.5"
     }
   }
 }

www/pnpm-lock.yaml

@@ -8,6 +8,8 @@ overrides:
   minimatch@>=5.0.0 <5.1.8: 5.1.8
   js-yaml@<4.1.1: 4.1.1
   webpack: 5.105.3
+  serialize-javascript: 7.0.4
+  immutable: 5.1.5
 
 importers:
   .:
@@ -760,6 +762,7 @@ packages:
       }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-libvips-linux-arm@1.2.4":
     resolution:
@@ -768,6 +771,7 @@ packages:
       }
     cpu: [arm]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-libvips-linux-ppc64@1.2.4":
     resolution:
@@ -776,6 +780,7 @@ packages:
       }
     cpu: [ppc64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-libvips-linux-riscv64@1.2.4":
     resolution:
@@ -784,6 +789,7 @@ packages:
       }
     cpu: [riscv64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-libvips-linux-s390x@1.2.4":
     resolution:
@@ -792,6 +798,7 @@ packages:
       }
     cpu: [s390x]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-libvips-linux-x64@1.2.4":
     resolution:
@@ -800,6 +807,7 @@ packages:
       }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-libvips-linuxmusl-arm64@1.2.4":
     resolution:
@@ -808,6 +816,7 @@ packages:
       }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   "@img/sharp-libvips-linuxmusl-x64@1.2.4":
     resolution:
@@ -816,6 +825,7 @@ packages:
       }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   "@img/sharp-linux-arm64@0.34.5":
     resolution:
@@ -825,6 +835,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-linux-arm@0.34.5":
     resolution:
@@ -834,6 +845,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [arm]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-linux-ppc64@0.34.5":
     resolution:
@@ -843,6 +855,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [ppc64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-linux-riscv64@0.34.5":
     resolution:
@@ -852,6 +865,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [riscv64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-linux-s390x@0.34.5":
     resolution:
@@ -861,6 +875,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [s390x]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-linux-x64@0.34.5":
     resolution:
@@ -870,6 +885,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   "@img/sharp-linuxmusl-arm64@0.34.5":
     resolution:
@@ -879,6 +895,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   "@img/sharp-linuxmusl-x64@0.34.5":
     resolution:
@@ -888,6 +905,7 @@ packages:
     engines: { node: ^18.17.0 || ^20.3.0 || >=21.0.0 }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   "@img/sharp-wasm32@0.34.5":
     resolution:
@@ -1194,6 +1212,7 @@ packages:
     engines: { node: ">= 10" }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   "@next/swc-linux-arm64-musl@16.1.6":
     resolution:
@@ -1203,6 +1222,7 @@ packages:
     engines: { node: ">= 10" }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   "@next/swc-linux-x64-gnu@16.1.6":
     resolution:
@@ -1212,6 +1232,7 @@ packages:
     engines: { node: ">= 10" }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   "@next/swc-linux-x64-musl@16.1.6":
     resolution:
@@ -1221,6 +1242,7 @@ packages:
     engines: { node: ">= 10" }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   "@next/swc-win32-arm64-msvc@16.1.6":
     resolution:
@@ -1645,6 +1667,7 @@ packages:
     engines: { node: ">= 10.0.0" }
     cpu: [arm]
     os: [linux]
+    libc: [glibc]
 
   "@parcel/watcher-linux-arm-musl@2.5.6":
     resolution:
@@ -1654,6 +1677,7 @@ packages:
     engines: { node: ">= 10.0.0" }
     cpu: [arm]
     os: [linux]
+    libc: [musl]
 
   "@parcel/watcher-linux-arm64-glibc@2.5.6":
     resolution:
@@ -1663,6 +1687,7 @@ packages:
     engines: { node: ">= 10.0.0" }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   "@parcel/watcher-linux-arm64-musl@2.5.6":
     resolution:
@@ -1672,6 +1697,7 @@ packages:
     engines: { node: ">= 10.0.0" }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   "@parcel/watcher-linux-x64-glibc@2.5.6":
     resolution:
@@ -1681,6 +1707,7 @@ packages:
     engines: { node: ">= 10.0.0" }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   "@parcel/watcher-linux-x64-musl@2.5.6":
     resolution:
@@ -1690,6 +1717,7 @@ packages:
     engines: { node: ">= 10.0.0" }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   "@parcel/watcher-win32-arm64@2.5.6":
     resolution:
@@ -2143,6 +2171,7 @@ packages:
       }
     cpu: [arm]
     os: [linux]
+    libc: [glibc]
 
   "@rollup/rollup-linux-arm-musleabihf@4.59.0":
     resolution:
@@ -2151,6 +2180,7 @@ packages:
       }
     cpu: [arm]
     os: [linux]
+    libc: [musl]
 
   "@rollup/rollup-linux-arm64-gnu@4.59.0":
     resolution:
@@ -2159,6 +2189,7 @@ packages:
       }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   "@rollup/rollup-linux-arm64-musl@4.59.0":
     resolution:
@@ -2167,6 +2198,7 @@ packages:
       }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   "@rollup/rollup-linux-loong64-gnu@4.59.0":
     resolution:
@@ -2175,6 +2207,7 @@ packages:
       }
     cpu: [loong64]
     os: [linux]
+    libc: [glibc]
 
   "@rollup/rollup-linux-loong64-musl@4.59.0":
     resolution:
@@ -2183,6 +2216,7 @@ packages:
       }
     cpu: [loong64]
     os: [linux]
+    libc: [musl]
 
   "@rollup/rollup-linux-ppc64-gnu@4.59.0":
     resolution:
@@ -2191,6 +2225,7 @@ packages:
       }
     cpu: [ppc64]
     os: [linux]
+    libc: [glibc]
 
   "@rollup/rollup-linux-ppc64-musl@4.59.0":
     resolution:
@@ -2199,6 +2234,7 @@ packages:
       }
     cpu: [ppc64]
     os: [linux]
+    libc: [musl]
 
   "@rollup/rollup-linux-riscv64-gnu@4.59.0":
     resolution:
@@ -2207,6 +2243,7 @@ packages:
       }
     cpu: [riscv64]
     os: [linux]
+    libc: [glibc]
 
   "@rollup/rollup-linux-riscv64-musl@4.59.0":
     resolution:
@@ -2215,6 +2252,7 @@ packages:
       }
     cpu: [riscv64]
     os: [linux]
+    libc: [musl]
 
   "@rollup/rollup-linux-s390x-gnu@4.59.0":
     resolution:
@@ -2223,6 +2261,7 @@ packages:
       }
     cpu: [s390x]
     os: [linux]
+    libc: [glibc]
 
   "@rollup/rollup-linux-x64-gnu@4.59.0":
     resolution:
@@ -2231,6 +2270,7 @@ packages:
       }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   "@rollup/rollup-linux-x64-musl@4.59.0":
     resolution:
@@ -2239,6 +2279,7 @@ packages:
       }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   "@rollup/rollup-openbsd-x64@4.59.0":
     resolution:
@@ -2668,6 +2709,7 @@ packages:
     engines: { node: ">= 20" }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   "@tailwindcss/oxide-linux-arm64-musl@4.2.1":
     resolution:
@@ -2677,6 +2719,7 @@ packages:
     engines: { node: ">= 20" }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   "@tailwindcss/oxide-linux-x64-gnu@4.2.1":
     resolution:
@@ -2686,6 +2729,7 @@ packages:
     engines: { node: ">= 20" }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   "@tailwindcss/oxide-linux-x64-musl@4.2.1":
     resolution:
@@ -2695,6 +2739,7 @@ packages:
     engines: { node: ">= 20" }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   "@tailwindcss/oxide-wasm32-wasi@4.2.1":
     resolution:
@@ -3162,6 +3207,7 @@ packages:
       }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   "@unrs/resolver-binding-linux-arm64-musl@1.11.1":
     resolution:
@@ -3170,6 +3216,7 @@ packages:
       }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   "@unrs/resolver-binding-linux-ppc64-gnu@1.11.1":
     resolution:
@@ -3178,6 +3225,7 @@ packages:
       }
     cpu: [ppc64]
     os: [linux]
+    libc: [glibc]
 
   "@unrs/resolver-binding-linux-riscv64-gnu@1.11.1":
     resolution:
@@ -3186,6 +3234,7 @@ packages:
       }
     cpu: [riscv64]
     os: [linux]
+    libc: [glibc]
 
   "@unrs/resolver-binding-linux-riscv64-musl@1.11.1":
     resolution:
@@ -3194,6 +3243,7 @@ packages:
       }
     cpu: [riscv64]
     os: [linux]
+    libc: [musl]
 
   "@unrs/resolver-binding-linux-s390x-gnu@1.11.1":
     resolution:
@@ -3202,6 +3252,7 @@ packages:
       }
     cpu: [s390x]
     os: [linux]
+    libc: [glibc]
 
   "@unrs/resolver-binding-linux-x64-gnu@1.11.1":
     resolution:
@@ -3210,6 +3261,7 @@ packages:
       }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   "@unrs/resolver-binding-linux-x64-musl@1.11.1":
     resolution:
@@ -3218,6 +3270,7 @@ packages:
       }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   "@unrs/resolver-binding-wasm32-wasi@1.11.1":
     resolution:
@@ -5602,10 +5655,10 @@ packages:
         integrity: sha512-XREFCPo6ksxVzP4E0ekD5aMdf8WMwmdNaz6vuvxgI40UaEiu6q3p8X52aU6GdyvLY3XXX/8R7JOTXStz/nBbRw==,
       }
 
-  immutable@5.1.4:
+  immutable@5.1.5:
     resolution:
       {
-        integrity: sha512-p6u1bG3YSnINT5RQmx/yRZBpenIl30kVxkTLDyHLIMk0gict704Q9n+thfDI7lTRm9vXdDYutVzXhzcThxTnXA==,
+        integrity: sha512-t7xcm2siw+hlUM68I+UEOK+z84RzmN59as9DZ7P1l0994DKUWV7UXBMQZVxaoMSRQ+PBZbHCOoBt7a2wxOMt+A==,
       }
 
   import-fresh@3.3.1:
@@ -6402,6 +6455,7 @@ packages:
     engines: { node: ">= 12.0.0" }
     cpu: [arm64]
     os: [linux]
+    libc: [glibc]
 
   lightningcss-linux-arm64-musl@1.31.1:
     resolution:
@@ -6411,6 +6465,7 @@ packages:
     engines: { node: ">= 12.0.0" }
     cpu: [arm64]
     os: [linux]
+    libc: [musl]
 
   lightningcss-linux-x64-gnu@1.31.1:
     resolution:
@@ -6420,6 +6475,7 @@ packages:
     engines: { node: ">= 12.0.0" }
     cpu: [x64]
     os: [linux]
+    libc: [glibc]
 
   lightningcss-linux-x64-musl@1.31.1:
     resolution:
@@ -6429,6 +6485,7 @@ packages:
     engines: { node: ">= 12.0.0" }
     cpu: [x64]
     os: [linux]
+    libc: [musl]
 
   lightningcss-win32-arm64-msvc@1.31.1:
     resolution:
@@ -7875,11 +7932,12 @@ packages:
     engines: { node: ">=10" }
     hasBin: true
 
-  serialize-javascript@6.0.2:
+  serialize-javascript@7.0.4:
     resolution:
       {
-        integrity: sha512-Saa1xPByTTq2gdeFZYLLo+RFE35NHZkAbqZeWNd3BpzppeVisAqpDjcp8dyf6uIvEqJRd46jemmyA4iFIeVk8g==,
+        integrity: sha512-DuGdB+Po43Q5Jxwpzt1lhyFSYKryqoNjQSA9M92tyw0lyHIOur+XCalOUe0KTJpyqzT8+fQ5A0Jf7vCx/NKmIg==,
       }
+    engines: { node: ">=20.0.0" }
 
   set-function-length@1.2.2:
     resolution:
@@ -12920,7 +12978,7 @@ snapshots:
 
   immer@11.1.4: {}
 
-  immutable@5.1.4: {}
+  immutable@5.1.5: {}
 
   import-fresh@3.3.1:
     dependencies:
@@ -14539,7 +14597,7 @@ snapshots:
   sass@1.97.3:
     dependencies:
       chokidar: 4.0.3
-      immutable: 5.1.4
+      immutable: 5.1.5
       source-map-js: 1.2.1
     optionalDependencies:
       "@parcel/watcher": 2.5.6
@@ -14561,9 +14619,7 @@ snapshots:
 
   semver@7.7.4: {}
 
-  serialize-javascript@6.0.2:
+  serialize-javascript@7.0.4: {}
-    dependencies:
-      randombytes: 2.1.0
 
   set-function-length@1.2.2:
     dependencies:
@@ -14861,7 +14917,7 @@ snapshots:
       "@jridgewell/trace-mapping": 0.3.31
       jest-worker: 27.5.1
       schema-utils: 4.3.3
-      serialize-javascript: 6.0.2
+      serialize-javascript: 7.0.4
       terser: 5.46.0
       webpack: 5.105.3
 

www/public/service-worker.js

@@ -1,5 +1,13 @@
 let authToken = null;
 
+self.addEventListener("install", () => {
+  self.skipWaiting();
+});
+
+self.addEventListener("activate", (event) => {
+  event.waitUntil(self.clients.claim());
+});
+
 self.addEventListener("message", (event) => {
   if (event.data && event.data.type === "SET_AUTH_TOKEN") {
     authToken = event.data.token;
@@ -7,8 +15,8 @@ self.addEventListener("message", (event) => {
 });
 
 self.addEventListener("fetch", function (event) {
-  // Check if the request is for a media file
+  // Check if the request is for a media file (allow optional query params)
-  if (/\/v1\/transcripts\/.*\/audio\/mp3$/.test(event.request.url)) {
+  if (/\/v1\/transcripts\/.*\/audio\/mp3(\?|$)/.test(event.request.url)) {
     // Modify the request to add the Authorization header
     const modifiedHeaders = new Headers(event.request.headers);
     if (authToken) {