Skip to content
SASY

SASY

Policy enforcement for AI agents. Every tool call is checked against your Datalog policy and the history that led to it, before it runs.
Get startedView on GitHub

Your agent records each message and tool result in a dependency graph, and asks the SASY engine about each proposed tool call before it runs; the engine allows it or returns the denial reason to the agent. For LangChain, Google ADK and Langroid, sasy.instrument() does this automatically. Any other agent makes the same two calls with SASY’s primitives: sasy.record for each message and sasy.check_tool_call before each tool runs. Policies can follow influence through summaries, tools and interactions between agents, so you can control information flow, require approvals and enforce business rules outside the model.

SASY instrumentation connects an agent to its tools and services, records messages and dependencies in the engine's dependency graph, and checks proposed actions against developer policies. Allowed calls execute; denied calls return a reason to the agent.
SASY instrumentation records inputs, checks actions with the policy engine, and dispatches allowed calls. The policy engine evaluates developer rules over the action and its relevant history.