Let VocaLinux hit this host as an OpenAI transcription server

[jatinkrmalik]

Closes #36.

Desktop embed stays Planned. Phone pairing and the session API are untouched.

POST /v1/audio/transcriptions accepts the OpenAI multipart shape (file plus optional model, language, response_format, and stream). Auth is the same bearer token. Transcription goes through transcribe_adhoc, the same path the WebUI test recorder uses. The JSON body is {"text"}. The model field is ignored, so whisper-1 and sensevoice both hit whatever engine is loaded in the WebUI. language falls back to auto. verbose_json and stream=true return 400.

There is no /inference route.

Known limits: VocaLinux still defaults to /inference, so the dropdown has to be switched. Test Connection is GET / and does not prove the token.

[netlify]

✅ Deploy Preview for voca-gateway canceled.

Name	Link
🔨 Latest commit	38e9465
🔍 Latest deploy log	https://app.netlify.com/projects/voca-gateway/deploys/6a952e3f08572200086a8c0d

[greptile-apps]

Greptile Summary

The PR adds an authenticated OpenAI-compatible transcription endpoint for VocaLinux while preserving the existing engine and phone-session behavior.

• Registers POST /v1/audio/transcriptions and returns the OpenAI-compatible {"text": ...} response.
• Validates authentication, request length, audio type, language, response format, and streaming options.
• Routes uploaded audio through the existing ad-hoc transcription service and cleans up temporary files.
• Documents VocaLinux remote_api configuration and adds endpoint coverage.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
app/routes/transcriptions.py	Adds the authenticated OpenAI-compatible route with early request-size enforcement, bounded copying, validation, transcription, and cancellation-safe cleanup.
app/main.py	Registers the new transcription router without changing the existing session or streaming routes.
app/schemas.py	Adds the minimal OpenAI-compatible transcription response schema.
tests/test_transcriptions.py	Covers authentication, successful transcription, validation failures, upload limits, and early Content-Length handling.
README.md	Documents configuring VocaLinux to use the new OpenAI-compatible endpoint.
docs/configuration.md	Adds operational details and limitations for VocaLinux remote API use.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[VocaLinux POST multipart audio] --> B[Authenticate bearer token]
    B --> C[Validate Content-Length before multipart parsing]
    C --> D[Parse and validate form fields]
    D --> E[Copy audio with file-size limit]
    E --> F[Finalize temporary upload]
    F --> G[transcribe_adhoc using loaded WebUI engine]
    G --> H[Delete finalized audio]
    H --> I[Return JSON text]

Loading

_{Reviews (3): Last reviewed commit: "Require Content-Length on the OpenAI tra..." | Re-trigger Greptile}

[jatinkrmalik]

Shipping this. A running VocaLinux can point remote_api at POST /v1/audio/transcriptions on a self-hosted gateway. Desktop embed stays Planned.