ACL helps existing app capabilities (APIs, MCP tools, Zapier actions, CLIs) become safer for AI agents to use. You define the allowed actions and checks, ACL runs the workflow, and every run produces a receipt.
Use the guided tutorial in /agenticflow. Run a request, inspect the generated ACL, then open the receipt summary and full JSON proof.
Use the playground to load a template, modify the steps, and run it directly. Start with read flows, then try preview/confirm patterns for writes.
Expose a few actions as tools (API wrappers, MCP, Zapier, or CLI), then define one or two ACL flows around them with checks and receipts.
Run locally and iterate quickly:
Production recommendation: use the LLM to extract structured intent (JSON), compile to ACL templates, and keep freeform ACL generation as fallback for novel requests.