Voice Interface is a platform for building voice and chat AI agents that hold natural conversations, answer from your own knowledge, and take real actions through tools.
What you can build
- Voice agents that listen and speak in real time.
- Chat agents for text conversations.
- Agents that call tools — book calendar events, append to Google Sheets, hit your own APIs, or connect to MCP servers.
- Agents grounded in your documents — upload them as Resources and expose them through a Query tool.
Two ways to talk, one configuration
Every agent works over both channels from the same setup:
- Voice — a realtime speech-to-speech session. The model listens, thinks, and replies out loud.
- Chat — a text turn loop. You send messages; the model replies and may call tools along the way.
The system prompt, scope, tools, and knowledge you configure apply identically to both.
Core building blocks
| Building block | What it does |
|---|---|
| System prompt | The persona and instructions that steer every reply. |
| First message | What the agent says to open the conversation. |
| Models | A chat model and a voice model per agent — Gemini, OpenAI, or Nova. |
| Tools | Actions the agent can take mid-conversation. |
| Knowledge base | Documents the agent can search and cite. |
| Scope | strict (tools + knowledge only) or open (also general knowledge). |
TIP
New here? Head to the Quickstart to create and talk to your first agent in a few minutes.Where things live
- Dashboard (
/console) — the studio where you create and configure agents, tools, and Resources. - API Reference — every HTTP and realtime endpoint, for automating the platform.
Next steps
- How it works — what actually happens during a conversation.
- Quickstart — build your first agent.
- Healthcare agent — a complete, knowledge-grounded example.