Transcribe Call Import

Fan out diarization tasks for many rows in a single call.

Returns a summary with how many rows were queued and how many were skipped (broken down by reason) so the UI can show a meaningful toast even when nothing actually got enqueued (e.g. "All 12 rows already have transcripts").

Authorization

AuthorizationBearer <token>

JWT access token. Use: Authorization: Bearer .

In: header

Path Parameters

call_import_id*Call Import Id

Formatuuid

Header Parameters

Authorization?string|null

X-API-Key?string|null

X-EFFICIENTAI-API-KEY?string|null

X-Workspace-Id?string|null

Request Body

application/json

TypeScript Definitions

Use the request body type in TypeScript.

Body for kicking off diarization for one or many call-import rows.

The same shape powers both the per-row endpoint (where row_ids is ignored) and the batch-level endpoint. only_missing is the safe default G�� rows with an existing transcript are skipped unless overwrite_existing is set.

Two modes are supported:

mode="stt_llm" (default) G�� the legacy two-stage pipeline: STT produces plain text, an LLM splits it into agent/user turns using diarization_prompt. stt_provider and stt_model are required in this mode.
mode="llm_only" G�� skip STT entirely and hand the recording's audio bytes to a multimodal chat model along with diarization_prompt. The model both transcribes and diarises in a single pass. The STT fields are ignored (and must be omitted / null). Only providers whose chat API accepts audio input (OpenAI gpt-4o-audio-*, Google Gemini 1.5/2.0) are usable; other providers will surface a typed error on the row.

mode?Mode

Pipeline shape. 'stt_llm' (default) runs STT then an LLM diariser over the resulting text. 'llm_only' skips STT and feeds the raw audio to a multimodal LLM together with diarization_prompt for a single-pass transcribe + diarise.

Default"stt_llm"

Value in

"stt_llm"
"llm_only"

stt_provider?|

STT provider key, e.g. 'deepgram' or 'openai'. Required when mode='stt_llm'; must be omitted when mode='llm_only'.

stt_model?|

STT model name, e.g. 'nova-2' or 'whisper-1'. Required when mode='stt_llm'; must be omitted when mode='llm_only'.

credential_id?|

Optional AIProvider/Integration row to pin for this run.

language?|

Optional ISO language hint, e.g. 'en'.

only_missing?Only Missing

When true, rows with an existing transcript are skipped (the default safe behavior).

Defaulttrue

overwrite_existing?Overwrite Existing

When true, existing transcripts are replaced. Mutually exclusive with only_missing.

Defaultfalse

row_ids?array<>|

Restrict the run to a specific subset of rows. NULL = every row in the import (subject to only_missing).

diarization_llm_provider*Diarization Llm Provider

LLM provider that diarises the call. In stt_llm it sees the STT text; in llm_only it sees the raw audio.

Lengthlength <= 50

diarization_llm_model*Diarization Llm Model

LLM model name. In llm_only mode this must be a model that accepts audio input (e.g. 'gpt-4o-audio-preview', 'gemini-1.5-pro').

Lengthlength <= 100

diarization_llm_credential_id?|

Optional AIProvider row to pin for the diarisation LLM.

diarization_prompt?|

Operator-supplied system prompt for the diariser LLM. When NULL/empty the worker uses the canonical default (see GET /api/v1/call-imports/diarisation-prompt-default).

Response Body

`application/json`

curl -X POST "https://example.com/api/v1/call-imports/497f6eca-6276-4993-bfeb-53cbbbba6f08/transcribe" \  -H "Content-Type: application/json" \  -d '{    "diarization_llm_provider": "string",    "diarization_llm_model": "string"  }'

{  "queued": 0,  "skipped_rows": 0,  "skipped_reason_counts": {    "property1": 0,    "property2": 0  },  "accepted": false}

Community & contact

Found a bug or have a feature request? Open a GitHub issue.
Join our Discord for faster replies!

Authorization

Path Parameters

Header Parameters

Request Body

Response Body

202application/json

422application/json

Community & contact

`application/json`

`application/json`