Open-source permissions for AI agents that delegate: each agent task gets the permissions it needs, derived from your app; a sub-agent never holds more than its parent; every decision lands in an audit log you verify offline, with us absent.
- attenu-guard — the enforcement library. Zero-dependency core, 12 frameworks unmodified (LangGraph, CrewAI, Google ADK, OpenAI Agents, Claude Agent SDK, MCP, …).
pip install attenu-guard - attenu-derive — the engine that computes the permission set from the app itself.
pip install attenu-derive - Docs: https://attenu.io/docs/ · Recipes: https://attenu.io/blog/ · Talk to us: https://attenu.io/contact/
Questions and integration reports: Discussions. Vulnerabilities: the private advisory form on either repository, never a public issue. Contributions welcome — read CONTRIBUTING.md (DCO, Apache-2.0).
