Endpoints

Speech intelligence

Ask questions grounded on a transcript and run chat completions over your voice data.

Two endpoints turn finished transcripts into answers. POST /v2/speech-understanding answers a single prompt strictly from one transcript. POST /v2/llm/chat-completions runs a multi-turn chat, optionally grounded on a transcript. Both require the referenced transcript to be completed (409 not_ready otherwise).

Speech understanding

POST/v2/speech-understandingAnswer one prompt strictly from a transcript

The model is instructed to answer only from the transcript text. When the transcript does not contain the answer, the response says so instead of guessing.

Request body

Field	Type	Default	Description
transcript_id	string	required	Id of a completed transcript created through the API.
prompt	string, 1 to 8000 chars	required	The question or instruction, for example `Summarize the action items`.
model	string or null	null	Optional reasoning model override. Leave unset to use the service default.
response_format	string	text	`text` for prose, `json` to have the model answer with a JSON object.

Response

Field	Type	Description
id	string	Id of this speech understanding result.
transcript_id	string	The transcript the answer is grounded on.
model	string	The model that produced the answer.
response	string	The answer. With `response_format` set to `json`, this string contains a JSON object to parse.
created_at	string	ISO 8601 creation time.

Errors

Status	Code	When
404	not_found	The transcript does not exist or is not yours.
409	not_ready	The transcript is not completed or has no text yet.

bash

curl -X POST https://api.realtimevoicekit.com/v2/speech-understanding \
  -H "Authorization: Bearer rtvk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "transcript_id": "tr_5c8e2f91d0",
    "prompt": "List the action items with owners.",
    "response_format": "text"
  }'

Chat completions

POST/v2/llm/chat-completionsMulti-turn chat, optionally grounded on a transcript

Request body

Field	Type	Default	Description
messages	array, 1 to 32 items	required	Conversation messages in order. Each message has `role` (`system`, `user`, or `assistant`) and `content` (1 to 24000 characters).
model	string or null	null	Optional reasoning model override.
transcript_id	string or null	null	When set, the transcript's text (first 60000 characters) is injected as source context so the chat can reference the recording.
response_format	string	text	`text` or `json`. With `json`, the assistant answers with a JSON object.

Response

Field	Type	Description
id	string	Id of this completion.
object	string	Always `chat.completion`.
created_at	string	ISO 8601 creation time.
model	string	The model that produced the completion.
choices	array	One choice: `index`, `message` with `role` (`assistant`) and `content`, and `finish_reason` (`stop`).

Errors

Status	Code	When
404	not_found	`transcript_id` was set but does not exist or is not yours.
409	not_ready	`transcript_id` was set but the transcript is not completed.
422	(validation)	Empty `messages`, a bad `role`, or `content` out of bounds.

bash

curl -X POST https://api.realtimevoicekit.com/v2/llm/chat-completions \
  -H "Authorization: Bearer rtvk_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "transcript_id": "tr_5c8e2f91d0",
    "messages": [
      {"role": "user", "content": "What pricing objections came up in this call?"}
    ]
  }'

Example response

{
  "id": "chat_1f6b09c2aa",
  "object": "chat.completion",
  "created_at": "2026-07-22T14:21:37Z",
  "model": "voicekit-chat",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "Two objections came up: annual pricing felt high, and the team asked about volume discounts."
      },
      "finish_reason": "stop"
    }
  ]
}