halyn

<div align="center">

AI agents act in the world. Halyn makes sure they can't break it.

[](https://pypi.org/project/halyn/) [](https://pypi.org/project/halyn/)

The safety layer for AI agents · Shield rules · Audit chain · Built-in dashboard

Website · Quick Start · Dashboard · Shield Rules

</div>

---

AI agents are acting in the real world — deleting files, restarting servers, controlling robots. Claude has 22 MCP tools. GPT has computer use.

But who stops them from breaking things?

Today the answer is: the prompt says "be careful." That's like telling a car "don't crash" instead of installing brakes.

Halyn is the brakes.

pip install halyn
halyn-mcp --port 8935
# Open http://localhost:8935 → see the dashboard

30 seconds. Zero code. Working dashboard.

Built-in web UI. Type commands, see shields, watch the audit chain update in real-time.

Constraints enforced at protocol level. The AI has no power to override them.

from halyn import ControlPlane, SSHDriver

cp = ControlPlane()
cp.connect(SSHDriver("192.168.1.10", "admin"))
cp.shield("deny * delete *")    # no deletions — ever

cp.act("restart nginx")   # ✓ allowed
cp.act("rm -rf /etc")     # ✗ blocked — always

Not guidelines. Physics.

Every action recorded in SHA-256 hash chain. Tamper-evident, append-only.

SSH, HTTP, WebSocket, Serial, MQTT, OPC-UA, ROS2, DDS, Docker, Browser, Unitree, Socket.

Halyn implements NRP — 6 rules that no AI can break.

---

halyn.dev · nrprotocol.dev · PyPI · [contact@halyn.dev](mailto:contact