Documents, audio & editing
Configure the document parser, conversion services, audio providers, and PDF export used by the workspace.
Document parsing and OCR
OCR_DEFAULT_PROVIDER selects the parser. Local Compose defaults to DOCLING, which needs --profile ocr. Cloud options include AZURE, LANDING_AI, and DATALAB; set the selected provider's credentials on both app and worker.
- AZURE
- AZURE_DOC_INTELLIGENCE_ENDPOINT + AZURE_DOC_INTELLIGENCE_KEY.
- LANDING_AI
- LANDING_AI_API_KEY.
- DATALAB
- DATALAB_API_KEY.
- DOCLING
- Local Docling service through the document converter; enable the ocr profile.
Use DOCUMENT_CONVERTER_URL for the current conversion service. OCR_WORKER_URL is obsolete; OCR_ROUTER_URL is a deprecated fallback.
Conversion, editing, and PDF export
Compose wires these services and local development credentials. For remote services, share the matching credential with each server and allow only the origins from which that service needs to fetch files.
- Document converter
- DOCUMENT_CONVERTER_URL + DOCUMENT_CONVERTER_API_KEY on clients. Server uses CONVERTER_API_KEY and ALLOWED_FETCH_ORIGINS; Compose reads CONVERTER_ALLOWED_FETCH_ORIGINS.
- ADEU editing
- ADEU_SERVICE_URL + ADEU_SERVICE_API_KEY. Configure ADEU_ALLOWED_FETCH_ORIGINS for app and storage origins.
- Gotenberg PDF export
- GOTENBERG_SERVICE_URL, GOTENBERG_SERVICE_USERNAME, and GOTENBERG_SERVICE_PASSWORD. Uses Basic auth.
Missing conversion services can surface as a typed 503 for the affected action. Health-check the service and its credentials before retrying a failed document.
Audio transcription
Cloud is the default transcription mode. Configure TRANSCRIPTION_API_BASE_URL + TRANSCRIPTION_API_KEY and TRANSCRIPTION_MODEL independently from chat when using a dedicated compatible endpoint.
For local Whisper, set TRANSCRIPTION_PROVIDER=sidecar plus TRANSCRIPTION_SERVICE_URL and TRANSCRIPTION_SERVICE_API_KEY. The service uses TRANSCRIPTION_API_KEY for the matching server credential. The production overlay requires --profile whisper, and TRANSCRIPTION_PROVIDER must be explicitly forwarded into app and worker.
Spoken responses
Text-to-speech uses GOOGLE_AI_API_KEY with Google Cloud Text-to-Speech enabled. GEMINI_TTS_VOICE selects the voice; the current default is en-US-Chirp3-HD-Kore. A chat endpoint key alone does not enable this service.