Endo Familiar
prototypeA JavaScript agent sandbox for secure AI agent execution
Details
Endo Familiar is a working prototype of an object-capability based JavaScript runtime for AI agents. It provides a secure environment where agents can operate with controlled permissions, running in sandboxed compartments with carefully scoped access. The framework ensures agents only have the minimum authority needed to perform their specific tasks.
Best fit users
- •Developers building AI agents
- •Organizations concerned with AI security
- •Teams working with critical infrastructure
Why this one made the cut
Most AI agent frameworks grant too much authority that could lead to catastrophic failures if exploited. Endo Familiar fundamentally changes this by implementing the 'Authority Follows References' model, restricting access so agents only have what they're explicitly granted. This prevents over-privileged operations and potential security breaches while still enabling useful agent capabilities.
What makes it different
Endo Familiar enforces the Principle of Least Authority through structural guarantees in its architecture, not just policy enforcement. It names capabilities for human readability and implements a pet name system to simplify and clarify capability management.