Model selection
Pick a function-calling model, effort, and fast mode for an Action Agent. Provider keys stay on the organization.
Model selection
The Action Agent card in Studio holds the system prompt and the model picker. Only models with function calling (tool use) appear. Credentials come from Organization → Integrations → AI Vendors, the same keys Actionflow Studio uses.
There is no API or SDK method that calls a model directly. Pick the model on the agent, then chat. For flow graphs, put an AI node on an ActionFlow and run it.
What you set
| Field | Purpose |
|---|---|
| System prompt | Base instructions. Keep it short. Put reusable policy in skills. |
| Provider / platform model | Models that support tools |
| Effort | Reasoning effort when the model supports it |
| Fast mode | Lower latency when the model supports it |
Edits apply on the next message after save (configVersion). Prompt and model changes are not hot-swapped mid-reply.
Choosing a model
Match quality, speed, cost, and capabilities to the job. Agents that call MCP or tools need reliable tool calling.
For provider-by-provider comparison, see the AI Model Selection Guide.