Skip to content

API.md

Latest commit

 

History

History
81 lines (66 loc) · 3.88 KB

API.md

File metadata and controls

81 lines (66 loc) · 3.88 KB
Raw

API

The alpha API serves one local user. Contracts may change before a stable release. Except for health, model catalog, login, and limited status endpoints, /v1/* requires authentication when local bypass is disabled.

Meetings

Method Path Purpose
POST `/v1/meetings/upload?mode=quick meeting`
GET /v1/meetings List meetings
GET /v1/meetings/search?q=... Search transcript text
GET/PATCH/DELETE /v1/meetings/{id} Read, rename, or delete a meeting
GET /v1/meetings/{id}/audio Stream managed audio
PATCH /v1/meetings/{id}/segments/{segment_id} Correct text
PATCH /v1/meetings/{id}/segments/speaker Assign selected segments
PATCH /v1/meetings/{id}/speakers/{speaker_id}/person Confirm or clear a person
POST/PUT `/v1/meetings/{id}/notes/{summary minutes
GET `/v1/meetings/{id}/export?format=markdown srt

status=ready means the transcript is usable. transcript_state=draft|refined distinguishes a provisional live transcript from a completed refinement. Check diarization_status separately: completed, unavailable, pending, or not_requested. A usable draft is retained if refinement fails. While status=processing, draft text, speaker/person assignments, and meeting outputs are read-only so the atomic refinement cannot overwrite user changes. Meeting detail includes the most recent processing_job for stage/progress UI.

Meeting detail may include a processing_manifest describing the immutable provenance of that processing run. Clients must treat its nested fields as optional for records created before manifests were introduced:

{
  "version": 1,
  "strategy": "external-diarization",
  "asr": {"engine": "qwen3", "model": "Qwen/Qwen3-ASR-0.6B", "timestamp_granularity": "word"},
  "diarization": {"provider": "pyannote", "status": "completed", "alignment": "word-timestamps"},
  "speaker_identity": {"engine": "campplus", "model_id": "modelscope:..."},
  "generated_at": "2026-07-16T10:00:00Z"
}

SPEAKER_00, SPEAKER_01, and similar labels are anonymous identifiers scoped to one meeting. They are not person IDs. identity_status is one of anonymous, suggested, auto_matched, or confirmed. Strict matches may be displayed and exported automatically with provenance; suggestions remain anonymous in the transcript until confirmed through the speaker/person PATCH.

Jobs and people

  • GET /v1/jobs, GET /v1/jobs/{id}
  • POST /v1/jobs/{id}/cancel, POST /v1/jobs/{id}/retry
  • GET/POST /v1/people
  • GET/PATCH/DELETE /v1/people/{id}
  • POST /v1/people/{id}/samples
  • GET /v1/people/{id}/samples/{sample_id}/audio
  • DELETE /v1/people/{id}/samples/{sample_id}

People are locally enrolled identities. Automatic display requires compatible embedding models, enough meeting speech, at least two strong enrollment samples, and stricter confidence/ambiguity thresholds than ordinary suggestions.

Runtime configuration

  • GET /v1/models, GET /v1/engines, PUT /v1/engine
  • GET /v1/asr/engines, PUT /v1/asr/engine
  • GET/PUT /v1/llm/settings, GET /v1/llm/status, POST /v1/llm/test
  • POST /v1/auth/login, GET /v1/auth/me, POST /v1/auth/change-password
  • GET /health, GET /ready
  • WS /ws/v1/stream/{client_id}

OpenAPI is available at /docs while the server is running.

GET /v1/llm/status is passive and returns only configuration plus the most recent explicit test result. Only POST /v1/llm/test sends a minimal ping request to the configured OpenAI-compatible /chat/completions endpoint. Saving LLM settings clears the previous test result but does not contact the LLM service.

LLM API keys are read from the LLM_API_KEY process environment. They are not accepted by the settings API, returned to the browser, or stored in SQLite.