Agents

Define agents (model, system prompt, tool access, and guardrails) and compose them into higher-order agents.

What’s coming

Pick the model, write the system prompt, and set limits per agent.

Grant each agent exactly the MCP tools its job needs, nothing more.

Let an agent delegate to other agents for multi-step reasoning.

Preview the work in progress

This surface is still being built. Follow along in the live preview; it changes often and isn’t final.