control-plane

Agent Control Plane — Community Edition

Part of Agent OS - Kernel-level governance for AI agents

Policy-based governance for autonomous AI agents.

The Agent Control Plane provides a governance layer that sits between your AI agent and the actions it performs. Define policies in YAML or Python and the control plane enforces them deterministically before any action executes.

Installation

pip install agent-control-plane

Quick Start

from agent_control_plane import AgentControlPlane

plane = AgentControlPlane()
plane.load_policy("policies.yaml")

result = await plane.execute(
    action="database_query",
    params={"query": "SELECT * FROM users"},
    agent_id="analyst-001"
)
# Safe queries execute; destructive queries are blocked by policy

Features

• Deterministic policy enforcement (YAML or Python)
• Permission management and resource quotas
• Sandboxed execution with rollback support
• Audit logging via SQLite-based Flight Recorder
• Multi-framework support (OpenAI, LangChain, MCP, A2A)

Documentation

See docs/ for guides and CONTRIBUTING.md for development setup.

License

MIT License - see LICENSE for details.