Agent Governance Toolkit

What is Agent Governance Toolkit?

Agent Governance Toolkit (AGT) is an open-source Microsoft project that puts deterministic controls around AI agents instead of relying on prompt instructions. Each tool call, outgoing message and delegation passes through application code and is checked against a YAML policy before it runs, so a denied action such as dropping a table cannot happen. Each decision is written to a tamper-evident audit trail, and agents get their own identities so a shared API key no longer hides who did what. The project ships as separate packages for the policy engine, agent mesh, runtime sandboxing with privilege rings, SRE tooling such as a kill switch and chaos testing, and OWASP compliance checks. It installs with pip for Python and also has SDKs for TypeScript, .NET, Rust and Go, plus plugins for Claude Code and GitHub Copilot CLI. It is free to use, and the README labels it a public preview that may have breaking changes before general availability.

Use cases

• Block destructive agent tool calls
• Require human approval for risky actions
• Audit every agent decision
• Scan prompts for injection weaknesses
• Check agents against OWASP guidance
• Find unregistered shadow AI agents

Who it's for

• AI agent developers
• Platform engineers
• Security teams
• Compliance teams

Why use Agent Governance Toolkit

• Enforcement happens in code before the model's intent reaches a tool, so safety does not depend on the model following a prompt.
• Separate agent identities make incident review possible when several agents would otherwise share one API key.
• Audit records capture the active policy, the request and the allow or deny reason, which gives auditors something concrete to inspect.
• Every layer is optional, so a team can start with govern() and add identity, sandboxing or SRE controls later.

Agent Governance Toolkit features

• YAML policies that allow, deny or require approval per action
• govern() wrapper that checks and logs each tool call
• Agent identity using SPIFFE, DID and mTLS
• Tamper-evident audit log of every decision
• Runtime sandboxing with four privilege rings
• Kill switch, SLO monitoring and chaos testing
• agt CLI for doctor, verify, lint-policy and red-team scan
• MCP security gateway with tool poisoning detection
• Shadow AI discovery across processes, configs and repos
• SDKs for Python, TypeScript, .NET, Rust and Go

Agent Governance Toolkit integrations

• Microsoft Agent Framework
• Semantic Kernel
• AutoGen
• LangGraph
• LangChain
• CrewAI
• OpenAI Agents SDK
• Claude Code
• Google ADK
• LlamaIndex
• Haystack
• Mastra
• Dify
• Azure AI Foundry
• GitHub Copilot CLI

How to get started with Agent Governance Toolkit

1. Install the package with pip install "agent-governance-toolkit[full]" (Python 3.11 or newer), or pick the SDK for your language.
2. Write a policy.yaml with rules that allow, deny or require approval for action types.
3. Wrap an existing tool with govern(my_tool, policy="policy.yaml") so each call is evaluated and logged.
4. Blocked calls raise GovernanceDenied with the rule that stopped them, and each decision lands in the audit trail.
5. Run agt verify and agt lint-policy in CI to check OWASP coverage and policy files.

Agent Governance Toolkit limitations

• Public preview: breaking changes may occur before general availability
• Python has the full stack; the other four language SDKs cover core governance only
• Quick-start imports need the [full] extra, since the base wheel installs only the compliance CLI

Agent Governance Toolkit pricing

Agent Governance Toolkit is open source, so you can use and self-host it for free.

Prices can change and may vary by country or exclude tax. Check the official pricing page.

Agent Governance Toolkit alternatives

The main Agent Governance Toolkit alternatives listed on Debuly are NeMo Guardrails, Guardrails AI, Lakera Guard, LightAgent, Speakeasy AI Control Plane and Agno. The table compares pricing and platforms; order is A–Z.

1. 

   Agno

   Python framework and AgentOS runtime for building agents, teams and workflows, then serving them as an API with a control plane.

   Developer platformsAutomation and agentsCoding
   Open sourceVisit Agno (official site)
2. 

   Guardrails AI

   Open-source framework that wraps LLM apps with validators for PII, hallucination and policy checks, plus Snowglobe simulation testing.

   Developer platformsCodingAutomation and agents
   Open sourceVisit Guardrails AI (official site)
3. 

   Lakera Guard

   Runtime AI guardrails API that screens LLM and agent traffic for prompt attacks, data leakage and unsafe content, with a free Community tier.

   Developer platformsAutomation and agents
   FreemiumVisit Lakera Guard (official site)
4. 

   LightAgent

   Compact open-source Python framework for building agents with tools, memory, MCP, tracing, guardrails and multi-step workflows.

   Developer platformsAutomation and agentsCoding
   Open sourceVisit LightAgent (official site)
5. 

   NeMo Guardrails

   Open-source Python library from NVIDIA that adds programmable safety, topic and security rails between your app and an LLM.

   Developer platformsAutomation and agents
   Open sourceVisit NeMo Guardrails (official site)
6. 

   Speakeasy AI Control Plane

   Governance layer between AI agents and company systems that connects MCP servers, filters risky data, scopes access by role and logs every tool call.

   Developer platformsAutomation and agents
   PaidVisit Speakeasy AI Control Plane (official site)

Agent Governance Toolkit FAQ

Sources & verification

Facts on this page come from the official sources below, each with the date we read it. Written by Debuly; the vendor did not review this page.

Image credits

• Logo: from avatars.githubusercontent.com · fetched 24 Oct 2026 · Vendor trademark, used only to identify the product; removed on request
• Cover: none collected yet, so a generated tile is shown.

Spot an error? Report an issue with this listing. We also remove a logo or image on the owner's request.