Chat

The same agent runs as text. The prompt, the tools and the knowledge base all behave the same - the difference is only that nothing is spoken.

This is useful for two things:

• Testing - iterate on a prompt in the Playground without talking out loud.
• A text channel - drive a chat from your own product over the API.

Over the API

A text session is opened against an agent, then driven a message at a time. Each turn returns what the agent said; the session's detail carries the whole conversation. See the API reference.

Voice-only settings (voice, transcriber, turn-taking) simply do not apply. Everything else - variables, extraction, tools, knowledge - is identical.