Build with Rexa.ai
A voice agent is three moving parts: something that starts the call, a prompt that decides what the agent says, and a stream of events that tells you what happened. The guides below follow that order.
If you are new, start with the quickstart — a free browser test room needs no phone number and no billing, so you can hear an agent before you connect anything. If you are integrating, jump to outbound calls and webhooks; between them they cover most of what an integration touches.
— Start here
Get a key, then get a call. About fifteen minutes end to end.
Quickstart
Signup to a live agent: create an API key, talk to the agent in a browser using a free test room, then place a real phone call from a connected number.
Authentication
The key format and its environments, the Bearer header every request carries, the scope required per endpoint, and how to rotate a key without downtime.
— API reference
One page per surface, each transcribed from the contract rather than summarised.
Outbound calls
POST /v1/calls field by field: the three required fields, the Idempotency-Key header, every session status value, the error codes, and how to poll a call to completion.
WebRTC rooms
Put an agent in a browser tab. Covers POST /v1/rooms and the free POST /v1/rooms/test, join tokens, room TTL, the test-room limits, and the error codes.
Functions
Register HTTPS tools the agent calls mid-conversation: JSON Schema parameters, timeouts, HMAC secret rotation, and the invocation log you debug them from.
Webhooks
The signed outbound envelope, how to verify the HMAC in Node or Python, retry semantics, and the full catalog of event types with the fields each one carries.
Telephony providers
Connect your own carrier account: the credential fields each of Telnyx, Twilio and Plivo needs, validation error codes, priority-based routing, failover, and the per-provider circuit breaker.
Compliance controls
What the platform enforces before and during a call: the pre-dispatch DNC gate, calling windows, two-party consent warnings, mid-call opt-out, and the audit trail it leaves.
Full API reference
Every endpoint, parameter, and response schema, generated from the OpenAPI contract and browsable in one screen. Opens in a new tab.
— SDKs
Typed clients for the two runtimes most integrations use. The REST API stays the source of truth.
SDK overview
Which clients exist, what each one wraps, and how to pick between calling the REST API directly and taking the dependency.
Node.js and TypeScript
Install the typed client, place a call, use the idempotency helper, and verify a webhook signature. Includes the real published package name.
Python
The same surface for Python 3.10 and up: client setup, placing calls, and verifying webhook signatures.
— The REST surface
Every public endpoint, so you can size the integration before reading a single guide. All of them take a Bearer API key and the scope named in the authentication guide.
| Method | Path | What it does | Guide |
|---|---|---|---|
| POST | /v1/calls | Dispatch an outbound phone call. | Outbound calls |
| POST | /v1/rooms | Provision a WebRTC room for browser-side voice. | WebRTC rooms |
| POST | /v1/rooms/test | Create a free, rate-limited test WebRTC room. | WebRTC rooms |
| GET | /v1/sessions | List recent sessions. | Outbound calls |
| GET | /v1/sessions/{id} | Retrieve a single session. | Outbound calls |
| GET | /v1/calls/{id}/recording | Download a call recording. | — |
| GET | /v1/voices | List the voices your key can use, stock and cloned. | Voice library |
| GET | /v1/credits/balance | Get the workspace credit balance. | — |
| GET | /v1/credits/transactions | List credit-ledger transactions. | — |
| GET | /v1/calls-control | Get the workspace calls-pause state. | — |
| POST | /v1/calls-control | Pause or resume all outbound calls. | — |
| GET | /v1/concurrency | Get the workspace concurrency settings. | — |
| POST | /v1/concurrency | Set or clear the self-imposed concurrency cap. | — |
— Trust
What the platform enforces, what you configure, and what stays your responsibility as the caller of record.
Security
Encryption in transit and at rest, access control, data retention windows, and the infrastructure posture behind the platform.
Compliance
Calling windows, DNC handling, recording consent, voice-clone consent, audit evidence, and which GDPR responsibilities sit with you as the caller of record.
Can’t find it?
The FAQ answers the product-level questions, the agent catalog shows what a configured agent looks like, and support will read a stack trace if it comes to that.