Packages¶

AGT publishes a small set of canonical package families. Most Python users should start with the toolkit; choose a narrower package only for a specific runtime surface or language SDK.

Public Preview

Package APIs may change before general availability. Use the canonical names below for new projects. Older component names remain as temporary compatibility packages.

Python toolkit¶

The meta-package is the recommended starting point. Its [full] extra installs the core runtime, framework integrations, CLI tools, and protocol packages used by the convenience-wrapper examples.

pip install "agent-governance-toolkit[full]"

Use a smaller extra when one framework is enough:

pip install "agent-governance-toolkit[langchain]"
pip install "agent-governance-toolkit[crewai]"
pip install "agent-governance-toolkit[openai-agents]"

Source: agent-governance-python/agent-compliance

Python core¶

agent-governance-toolkit-core contains policy, trust, identity, audit, and runtime primitives. It provides the current implementation behind the legacy agent_os, agentmesh, runtime, and hypervisor imports.

pip install agent-governance-toolkit-core

Source: agent-governance-python/agent-governance-toolkit-core

Framework integrations¶

agent-governance-toolkit-integrations contains optional adapters, so the core package does not install every framework dependency.

pip install "agent-governance-toolkit-integrations[langchain]"
pip install "agent-governance-toolkit-integrations[crewai,openai-agents]"

Available extras include LangChain, CrewAI, OpenAI Agents, LangGraph, LlamaIndex, Haystack, PydanticAI, Google ADK, Cedarling, and OpenShell.

Source: agent-governance-python/agent-governance-toolkit-integrations

CLI and operations¶

agent-governance-toolkit-cli contains operator commands, SRE and observability tooling, sandbox integrations, and MCP trust services.

pip install agent-governance-toolkit-cli

Source: agent-governance-python/agent-governance-toolkit-cli

Protocol governance¶

agent-governance-toolkit-protocols contains governance surfaces for MCP, A2A, trust protocols, and verifiable MCP receipts.

pip install agent-governance-toolkit-protocols

Source: agent-governance-python/agent-governance-toolkit-protocols

Agent Control Specification¶

Agent Control Specification, or ACS, is AGT's canonical stateless policy decision runtime. Python hosts use the native SDK to load manifests, build snapshots, call ACS, and enforce verdicts.

pip install agent-control-specification

agt-policies now ships the one-way agt migrate v4-to-v5 conversion tool. See the ACS package guide.

Language SDKs¶

SDK	Install	Source
TypeScript	`npm install @microsoft/agent-governance-sdk`	`agent-governance-typescript/`
.NET	`dotnet add package Microsoft.AgentGovernance`	`agent-governance-dotnet/`
Rust	`cargo add agentmesh`	`agent-governance-rust/agentmesh/`
Rust MCP	`cargo add agentmesh-mcp`	`agent-governance-rust/agentmesh-mcp/`
Go	`go get github.com/microsoft/agent-governance-toolkit/agent-governance-golang`	`agent-governance-golang/`

Other shipped surfaces¶

Some Python packages, developer tools, ACS SDKs, and OCI images remain separate where consolidation would obscure their lifecycle or intended consumer.

Status labels¶

Status	Meaning
Shipped	Released in a current package and covered by package-local validation.
Compatibility	Existing Microsoft-origin or legacy package identity retained temporarily.
Experimental	Runnable but not guaranteed stable.
Proposed	ADR/RFC/spec exists but implementation is not a shipped guarantee.
Vendor integration	Requires a vendor product, account, or platform.

Package pages should use these labels when a capability is not part of the canonical core release.

Compatibility and migration¶

Legacy distributions such as agent-os-kernel, agentmesh-platform, agentmesh-runtime, agent-sre, and agent-hypervisor are not recommended for new projects. Their compatibility pages remain available because existing imports and installations still need migration guidance.