Runtime access control layer for AI agents
Authed | Docs
Note: Authed Permissions is now the core focus of this repository. For identity and authentication, see Authed Identity.
As AI agents become more autonomous and interconnected, managing what they can access and do becomes critical. Authed Permissions offers a natural language approach to defining, enforcing, and managing permissions between agents and resources.
Write permission rules in an intuitive syntax that's both human-readable and machine-parsable:
GIVE READ ACCESS TO EMAILS WITH TAGS = WORK
DENY WRITE ACCESS TO ISSUES ASSIGNED TO = [email protected]
The permissions system adapts to virtually any resource type with custom fields and conditions, making it ideal for agents that need to access diverse data sources and APIs.
Easily add new data sources and APIs by defining integration mapping rules that translate between external schemas and permission statements.
Optimized for rapid permission checks without compromising security, ensuring agents can make quick decisions with proper authorization.
We'd love to hear from you! If you're:
- Building with MCP
- Building AI agents that need fine-grained access control
- Developing an agent building platform
- Creating integration tools that connect agents to external services
Visit getauthed.dev and book a call with our team.
Stay tuned for upcoming features including enhanced condition expressions, permission groups, role-based access control, audit logging, and performance optimizations.
MIT