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

Cerberus — an OpenAI-compatible AI policy gateway with identity-scoped aliases, free/paid routing policy, admin controls, and multi-model Fusion through OpenRouter under Cerberus policy..

Deterministic, rights-aware runtime and policy gateway for controlled AI content workflows, verifiable provenance, and reproducible accounting.

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.

Self-hosted TypeScript MCP gateway for ChatGPT and coding agents: scoped files, shell, Git, signed approvals, audit logs, and three security modes.

A small policy gateway for the Model Context Protocol: pinning, least privilege, egress allowlist, Rule of Two, approvals, audit. Companion code to the working paper "Who Decides" (A+43 Studios WP 2026-02).

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