The audio seconds an account may use each month: 1,800 (30 minutes) during the beta. Resets on the 1st, UTC.
The unit everything is counted in: the length of audio sent or returned. Text translation counts characters ÷ 15.
A secret starting with
namu_sk_ that identifies your account. Kept on your server, sent as Authorization: Bearer.Ready to build on with real users while quality keeps improving. Changes that could break you are announced in the changelog.
ha Hausa, dje Zarma, fr French: the ISO 639 codes the API uses.A compact audio format for speech. The API returns it with
"format": "opus" (Ogg Opus) and reads it in voice notes.Who answers in the voice agent.
nadia is the only one.429 quota_exceeded: the month’s allowance is used. Different from rate_limited, which passes after a few seconds.Requests per minute per key: 60 by default. Passing it returns
429 rate_limited with Retry-After.A unique id for every response, in
X-Request-Id and in error bodies. Send it to us when something goes wrong.A preview of work in progress, such as the voice agent. It can change without notice.
What a key may call:
transcribe, speech, translate, converse. New keys get all but converse.Who speaks in
/v1/speech: nadia (or warm), mouhamad, fr. Each is used with its speaker’s consent.