agent-governance-toolkit: a deterministic policy enforcement and governance layer for autonomous AI agents

What it solves

AI agents often operate autonomously, making decisions to call tools, browse the web, or query databases. Relying on prompt-level safety (asking the model to follow rules) is unreliable because models are susceptible to prompt injection and adversarial attacks. This toolkit provides a deterministic governance layer that intercepts agent actions before they are executed, ensuring that only allowed actions occur and providing a tamper-evident audit trail for compliance.

How it works

The toolkit implements a "fail-closed" architecture where every tool call, message, or delegation is intercepted by a policy engine. This engine evaluates the request against YAML-defined policies (using Rust-core runtimes for speed and determinism) and checks identity and trust levels. If an action is denied by the policy, it is structurally impossible for the agent to execute it, regardless of the model's intent.

Who it’s for

Developers and SREs deploying autonomous AI agents into production environments who need strict policy enforcement, identity management, and regulatory compliance (such as OWASP Agentic Top 10 or EU AI Act).

Highlights

  • Multi-Language Support: SDKs available for Python, TypeScript, .NET, Rust, and Go.
  • Framework Agnostic: Integrates with major frameworks like AutoGen, LangGraph, CrewAI, Semantic Kernel, and LlamaIndex.
  • Deterministic Enforcement: Moves safety from probabilistic prompt-based requests to deterministic application code.
  • Comprehensive Governance Stack: Includes tools for execution sandboxing (Privilege Rings), SRE monitoring (SLOs, kill switches), and compliance verification (OWASP audits).
  • MCP Security Gateway: Specifically detects tool poisoning, drift, and hidden instructions in Model Context Protocol servers.

Sources