A typed, policy-controlled, observable execution runtime for AI tool-using agents. Engineered to eliminate unconstrained side effects.
Every tool call is validated against strict Pydantic schemas before reaching executors. Unregistered tools fail fast at the boundary.
Built-in dry-run execution enables safe testing of planner intent without triggering mutable mutations in external CRMs or ERPs.
Every transition (Initiate ➔ Plan ➔ Validate ➔ Execute ➔ Complete) emits an immutable trace for observability and audit compliance.
// Test safety boundary with dry-run verification