Cervana AI Docs
The Cervana Voice API

Voice, as
infrastructure.

Speech-to-text, conversation, and speech synthesis behind one authenticated HTTPS endpoint — powered end-to-end by Cervana's own models, Percept, Noema, and Melo, with no third-party AI APIs. The same API runs in Cervana's cloud and in fully air-gapped on-premise deployments. Guides and the complete endpoint reference live in one place, generated from the versioned contract, so they can never drift from what's deployed.

Enter the documentation Make your first call
Base URL
https://api.cervanaai.com/v1
TLS 1.2/1.3 · plaintext requests are redirected.
Authentication
X-API-Key: cvn-…
Minted in the console (Settings → API keys), shown once.
Rate limits
per key · 429 + Retry-After
X-RateLimit-* headers on every authenticated response.
The voice pipeline, as an API
POST /transcribe · Cervana Percept 1.0

Speech to text

Caller audio in, finalized transcript out — streaming partials included.

 POST /chat · Cervana Noema 1.0

Chat completions

The language-model reasoning layer — streamed replies, OpenAI-compatible tool calling.

 POST /synthesize · Cervana Melo 1.0

Text to speech

Natural speech, streamed as WAV — preset or cloned voices.

 /voices

Voice cloning

Register consented voices, scoped strictly to your own key.

These are the raw building blocks. Phone agents themselves — greeting, voice, language, capabilities, number — are composed in the Cervana console, no code required; every call they answer runs on these same endpoints.

Start here
Guide

Quickstart

Mint a key, make your first calls, stream a chat reply.

 Guide

Tool Calling

Let your agent act mid-call — lookups, transfers, bookings.

 Guide

SIP Interconnect

Connect your trunk or PBX — everything your telephony team needs.

 Guide

Web Widget

One script tag puts your voice agent on any website.

 Reference

Every endpoint

Paths, schemas, error shapes, and a request sample for each.