Models & platforms

The model is an agent's brain. The Builder picks sensible defaults, and any agent's model can be changed at any time.

Choosing a model

The picker lives in an agent's editor drawer → Model & voice (Model & behaviour for text agents): a searchable list, filterable by type, each model shown with its provider. For voice agents the Language, Voice and Speech recognition choices cascade from the chosen model, so the model comes first. A swap autosaves into the draft like any other edit.

Temperature: precise or creative

Below the picker sits a temperature slider from Precise to Creative. Low settings make the agent consistent and script-faithful — right for confirmations and capture. Higher settings buy warmth and variety in open conversation, at the cost of predictability. A test call is the judge.

Agent types differ

Agents come in types with different capabilities:

• Phone agents answer real numbers and SIP registrations, speak with a chosen voice, and take keypad input.
• Browser agents take instant calls in a web page — the technology behind browser test calls.
• Text agents never take calls: they chat, and they make ideal subagents.

Some abilities — transfers, hand-backs, keypad tones — belong to particular agent types rather than to models. The Builder steers each design onto a type that supports the plays asked for.

The Workspace rule

Agents holding Google Workspace or Microsoft 365 tools — integration tools or calendar booking — may only run on approved model providers: OpenAI, Anthropic, Google and Ultravox. Providers without a citable commitment not to train on customer data are blocked from handling a connected account.

Enabling Workspace tools fails with a message naming any agent whose model doesn't qualify, and switching an already-armed agent onto a blocked provider is refused. The fix is the same either way: move that agent to an approved model.