Skip to main content
POST
Run a hybrid (semantic + keyword) search across an end user’s transcripts and get back the most relevant excerpts. Each result carries start_ms/end_ms so you can deep-link a citation back to the moment in the audio. This is the retrieval primitive behind cross-recording AI Ask. The standalone Node.js transcript-search example uses a configured end user and optional recording allowlist, then checks each returned recording’s ownership before printing excerpts. It retrieves citations without creating an Ask session or generating an answer. Search sends the query through the configured embedding provider path and may incur provider usage; empty results do not establish that transcription indexing succeeded.

Authentication

Requires an API key with recordings:read scope.

Request Body

string
required
Natural-language search query (1–1000 characters).
string
End user whose recordings to search (eu_*). Required unless the API key is itself scoped to an end user.
string[]
Restrict the search to these recordings. Omit to search all of the end user’s recordings.
string
Only include chunks from recordings on or after this time (ISO 8601).
string
Only include chunks from recordings before this time (ISO 8601).
integer
default:"8"
Maximum number of chunks to return (1–50).

Response

Returns the top-K most relevant transcript chunks, best first.

Response Fields

Search requires indexed transcript chunks. Automatic chunking and embedding after transcription depend on the effective processing.auto_embedding.enabled setting and an allowed embedding-provider route. A completed transcription alone does not prove indexing succeeded, and an empty search response is not an indexing status check.