agent-governance-toolkit vs shepherd
Agent governance and supervised execution compared
agent-governance-toolkit adds policy enforcement, identity, audit, and execution controls around agent actions. shepherd records agent runs and keeps workspace changes as proposals for review, replay, or application.
agent-governance-toolkit: Govern Autonomous AI Agents
Microsoft’s Agent Governance Toolkit uses the Agent Control Specification policy runtime to govern autonomous agents with policy enforcement, identity, sandboxing, and audit controls. It supports multiple agent frameworks and language SDKs, with Python providing the full stack.
shepherd: Supervise Agents with Reversible Execution Traces
Shepherd records agent work as inspectable, reversible execution traces and keeps changes as proposals for review. It is aimed at developers building systems that supervise, replay, or manage the work of other agents.
| agent-governance-toolkit | shepherd | |
|---|---|---|
| Language | Python | Python |
| License | MIT | MIT |
| Stars | 6.5k | 2.5k |
| Forks | 1.1k | 217 |
| Last analyzed | Oct 23, 2026 | Oct 4, 2026 |
Key differences
- agent-governance-toolkit governs actions such as tool calls and delegation; shepherd focuses on inspecting and managing the results of agent runs.
- agent-governance-toolkit uses policy decisions, identity, audit, and execution controls; shepherd uses traces, proposals, replay, and task-based repository permissions.
- agent-governance-toolkit integrates with several agent frameworks and offers SDKs in multiple languages; shepherd is a Python framework for building supervised agent systems.
- agent-governance-toolkit operates at the application middleware layer and recommends separate containers for OS-level isolation; shepherd enforces grants at the OS level on supported jailed devices.
- Both projects use the MIT license and are not archived; agent-governance-toolkit is Public Preview, while shepherd is in early alpha.
Choose agent-governance-toolkit if you…
- need policy checks, identity, or audit records for agent actions.
- want integrations with agent frameworks and SDKs across several languages.
- need controls such as approvals, sandboxing support, or compliance checks.
Choose shepherd if you…
- want to retain agent changes as proposals before applying them to a workspace.
- need to inspect, replay, or fork agent execution traces.
- are building systems to supervise, compare, or optimize other agents.
Compare the numbers side by side, or add more projects
This comparison is generated with AI from the OSRepos analyses of both projects. Always check each project's repository and documentation before choosing.