Transcripts

Send a call transcript for extraction and read back its status.

Note

A transcript is not a feedback item. It is read once, the customer's own words are extracted as verbatim quotes, and each quote becomes a feedback item with source call. The transcript itself is never stored. What is extracted, what is kept and how calls are counted is on the Sales calls page; this page is the contract.

POST/api/v1/transcriptssecret key

Send a transcript

One call per request. Returns 202 as soon as the row is recorded and the call is counted; extraction runs asynchronously, so poll GET /api/v1/transcripts/{id} for the result. Secret keys only: a widget key gets 401. 60 requests per minute per key. Bodies up to 2 MB.

Parameter	In	Description
`external_id`required	body	Your stable id for the call, 1 to 200 characters, such as the notetaker's meeting id. The idempotency key: a repeat returns 200 with the existing row and counts nothing.
`segments`	body	Exactly one of segments or text. An array of up to 5,000 { speaker, text, start_sec? }. speaker is "me", "them" or the label as written; text is one turn of speech, up to 20,000 characters; start_sec is seconds from the start of the call, 0 or more.
`text`	body	Exactly one of segments or text. The whole transcript as one string, read into at most 5,000 segments. With format "text" (default), lines such as Them: …, Me: …, [00:14:32] Them: …, Alice: … are read and an unlabelled line continues the previous segment. With format "vtt", WebVTT with optional <v Name> voice tags.
`format`	body	"text" (default) or "vtt". Only with text.
`title`	body	Up to 200 characters. Stored with personal names removed; company names stay.
`started_at`	body	ISO 8601 date-time. More than 48 hours before receipt makes the call history; missing counts as now, which is live. More than 24 hours in the future is refused.
`duration_sec`	body	Integer, 0 or more.
`language`	body	ISO 639-1 code of the customer's speech, for example "de".
`url`	body	https link back to the call in the source tool, up to 500 characters. Shown as the deep link on every signal.
`owner_email`	body	The person who ran the notetaker, the rep. Sets your own domain for the attendee filter; its local part also maps a named speaker to "me".
`owner_name`	body	How the rep is labelled in the transcript. A label equal to it (case-insensitive), or to its first name, maps to "me".
`folder`	body	Up to 100 characters. Your own taxonomy, kept as provenance.
`attendees`	body	Up to 50 { email?, name?, external? }. external defaults to "the email domain differs from owner_email's domain". Names are used to redact the transcript and are then dropped. Without at least one external attendee the call is recorded as skipped.

Speaker mapping: me and them are literal. Any other label equal (case-insensitive) to owner_name, to its first name, or to the local part of owner_email is "me"; every other label is "them". Text with no speaker labels at all gives every segment the speaker "unknown", and the call is recorded as skipped with reason no_quotable_speech: nothing is guessed.

Request (segments)

curl -X POST https://app.sentriment.com/api/v1/transcripts \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{
    "external_id": "note_8f3a2c",
    "title": "Acme discovery call",
    "started_at": "2026-09-12T14:00:00Z",
    "duration_sec": 2700,
    "url": "https://notes.example-notetaker.com/n/8f3a2c",
    "owner_email": "sam@yourcompany.example",
    "owner_name": "Sam Okafor",
    "folder": "Sales",
    "language": "en",
    "attendees": [
      { "email": "sam@yourcompany.example", "name": "Sam Okafor" },
      { "email": "jane@acme.example", "name": "Jane Doe" }
    ],
    "segments": [
      { "speaker": "me", "text": "Thanks for making the time. What prompted the call?", "start_sec": 12 },
      { "speaker": "them", "text": "Exports. Anything above 50k rows times out and the team exports every Monday.", "start_sec": 20 },
      { "speaker": "Jane Doe", "text": "Pricing per seat is hard to justify for the read-only users.", "start_sec": 69 }
    ]
  }'

Request (text)

curl -X POST https://app.sentriment.com/api/v1/transcripts \
  -H "Authorization: Bearer sk_live_…" \
  -H "Content-Type: application/json" \
  -d '{
    "external_id": "note_8f3a2c",
    "started_at": "2026-09-12T14:00:00Z",
    "owner_email": "sam@yourcompany.example",
    "attendees": [{ "email": "jane@acme.example" }],
    "format": "text",
    "text": "[00:00:12] Me: Thanks for making the time. What prompted the call?\n[00:00:20] Them: Exports. Anything above 50k rows times out."
  }'

202 Accepted · queued

{
  "id": "01KX8VPHE0XN0JCDM6GSRC5HDQ",
  "status": "queued",
  "metered_items": 6,
  "word_count": 2740
}

202 Accepted · skipped

{
  "id": "01KX8VPHE0XN0JCDM6GSRC5HDQ",
  "status": "skipped",
  "skip_reason": "no_quotable_speech",
  "metered_items": 0,
  "word_count": 912
}

200 OK · external_id already recorded

{
  "id": "01KX8VPHE0XN0JCDM6GSRC5HDQ",
  "status": "done",
  "metered_items": 6,
  "word_count": 2740
}

400 Bad Request

{
  "type": "https://sentriment.com/problems/invalid-request",
  "title": "Request body failed validation",
  "status": 400,
  "detail": "body: provide exactly one of segments or text"
}

Parameter	In	Description
`202`	status	New row. { id, status, metered_items, word_count, skip_reason? } with status queued, skipped or plan_blocked. metered_items is the count charged: one per 500 words, minimum 2.
`200`	status	external_id was already recorded in this project. { id, status, metered_items, word_count } for the existing row; nothing is counted again.
`400 invalid-request`	status	Validation failed; detail names the field, or the line count when the text could not be read or read into more than 5,000 segments.
`401 unauthorized`	status	Missing, revoked, or a widget (pk_) key.
`413 payload-too-large`	status	Body over 2 MB.
`429 rate-limited`	status	Over 60 per minute; Retry-After says how long.
`500 internal`	status	Nothing was recorded; safe to retry with the same external_id.

Skip reasons: too_short (under 800 words), too_long (over about three hours; a call is refused whole, never sliced), no_quotable_speech (no segment attributed to the customer), no_external_attendee and internal_call (the model read the call and judged it an internal one; set after the 202, so it is seen on GET). The first four are decided at the door before any reading and are not counted; internal_call is counted, because the reading was the cost.

GET/api/v1/transcripts/{id}secret key

Get one transcript

Status and counts for one call, scoped to the key's project. Poll it after a 202 to learn how many signals were found. No text is returned, because none is kept.

Parameter	In	Description
`id`required	path	The id from the 202 or 200 response.

Request

curl https://app.sentriment.com/api/v1/transcripts/01KX8VPHE0XN0JCDM6GSRC5HDQ \
  -H "Authorization: Bearer sk_live_…"

200 OK

{
  "id": "01KX8VPHE0XN0JCDM6GSRC5HDQ",
  "external_id": "note_8f3a2c",
  "provider": "api",
  "status": "done",
  "metered_items": 6,
  "word_count": 2740,
  "signal_count": 9,
  "created_at": "2026-09-12T15:02:11.000Z",
  "extracted_at": "2026-09-12T15:03:40.000Z"
}

Parameter	In	Description
`status`	field	queued · extracting · done · skipped · failed · plan_blocked.
`skip_reason`	field	Present when status is skipped.
`failure_reason`	field	Present when status is failed. A code, for example provider_5xx or schema_refused, never a message.
`signal_count`	field	Feedback items extracted from this call. 0 until status is done.
`extracted_at`	field	Present once extraction finished.
`404 not-found`	status	No such transcript in this project; another project's id is a 404, never a 403.

Was this page helpful?

← PreviousFeedback Next →Users & identity