SASY

SASY checks proposed agent actions against policies over messages and their dependencies. An integration records how messages flow through the application and asks the reference monitor for a decision before dispatching an action. Policies can use the action, authenticated identity and the context that produced it.

Start with a policy check, then follow the instrumentation contract to connect your framework. Learn how to express permissions and context-dependent denials in the policy language.

The configuration guide explains engine, SDK and session settings. Authentication covers API keys, identity and roles. Build and release describes the full Rust engine and the restricted guard runtime.

For Claude Code, use the integration at guard.sasy.ai, with local build guidance and management documentation here. Read current limits when deciding which guarantees your integration can rely on.