Skip to main content
POST
Create a transcription

Authorizations

Authorization
string
header
required

Workspace API key. Create one via the dashboard or POST /v1/auth/keys.

Query Parameters

async
enum<string>

Queue the transcription and return 202 with a request id instead of waiting. Recommended for long audio.

Available options:
true,
1

Body

input_audio
object
required

Base64 audio (JSON requests only). Mutually exclusive with the multipart file field.

model
string
default:auto

Exact owner/model slug, or auto / stt/auto for managed routing.

language
string

BCP-47 hint (e.g. es, pt-BR). On a managed route this skips language detection — lower latency and better accuracy.

prompt
string

Context or vocabulary hint for models that support prompting.

response_format
enum<string>
default:json
Available options:
json,
verbose_json,
text,
srt,
vtt,
diarized_json
temperature
number
timestamps
enum<string>
Available options:
none,
segment,
word,
both
diarize
boolean

Label speakers. Speaker ids are normalized to spk_N.

redact
boolean

Redact detected PII where supported.

translate
boolean

Translate the transcript to English where supported.

keyterms
string[]
retention
enum<string>
Available options:
none,
transcript,
debug,
archive
thinking
enum<string>
Available options:
minimal,
low,
medium,
high
metadata
object
provider
object

Per-request routing constraints, applied before ranking.

options
object

Native model controls, namespaced per owner. Applied only when the request resolves to that owner.

Response

The transcription. text/srt/vtt formats return a plain-text body instead.

id
string
object
string
Example:

"transcription"

created
integer
model
string

The model you requested.

resolved_model
string

The model that served the request. Managed routes report auto.

provider
string
text
string
language
string | null
duration
number | null

Audio duration in seconds.

segments
object[]
words
object[]
speakers
object[]
entities
object[]
usage
object

Normalized usage. cost is in USD; Naive credits are reported separately as credits_used.

route
object

How the model was selected, and what was attempted.

warnings
string[]
credits_used
number

Naive credits charged for this request.

credits_remaining
number