Voice Agent

Voice Agent

Add a hosted HUMAIN voice agent to your web or mobile app; your backend starts each call with its API key and your user talks to the agent over WebRTC.

Voice Agent is in beta.

The HUMAIN voice agent listens, answers, and speaks in Arabic and English with HUMAIN speech recognition and voices. HUMAIN hosts and scales it. You decide what it says (prompt, intro, voice) and what it can do in your app (client tools).

How a call works

One Voice Agent call
1. Start
Your backend

Sends its API key to the session endpoint and gets a short-lived token.

2. Join
Your user's browser or app

Connects with the token. The agent joins the same call.

3. Talk
Over WebRTC

Audio, live transcripts, and tool calls flow between the user and the agent.

4. End
Either side

The call ends with a reason you can show the user.

The API key stays on your server. The browser only receives a token that admits one user to one call; see Session options. Billing and plan limits are in Limits and billing.

Start sessions from your backend with @humain-voice/sdk (Node.js or Bun, import from @humain-voice/sdk/agent) or humain-voice (Python, import humain_voice.agent), or call the session endpoint from any language. Join from the browser with @humain-voice/agent, or from iOS, Android, Flutter, and React Native with LiveKit clients. Install lines are on the SDKs page.

On this page