policy-gateway

Open-source runtime security and governance for AI agents, MCP tools, robotics, industrial automation, and physical AI.

Agentic security control plane for MCP and AI agent tool calls. MCP-native policy gateway with topology discovery and audit.

Governed AI computers: policy, sandbox, gateway enforcement, and OpenVTC approval integration

MCP policy gateway (Python) for boundary-governed MCP tool execution. Intercepts tool calls, applies allow/deny/step-up approval rules, logs append-only evidence and supports proxy routing to downstream MCP servers. Includes local admin API, CLI approver workflow and sandboxed shell/fs tools for prototype evaluation and integration design.

AI Engineering Control Plane - portable, agent-agnostic control plane. v2.1.0: 52 ADRs, 42 skills, 15 workflows, 5 adapters, 1756 assertions. SWE-bench Verified Pass@1 adapter, first real live-LLM PASS proof, 3-tier sandbox, OpenTelemetry, visual regression, +98.6% JIT context savings, RuntimePolicyGateway & Merkle audit chain, subagent swarms.

Agent Tool Policy Gateway prototype: signed grants, sandboxed ToolCall, replayable evidence for multi-Agent software repair.