halyn 0.3.3

Halyn — NRP control plane with domain-scoped authorization.

Halyn

Connect any device to any AI.

MCP Server · 12 drivers · Shield rules · Audit chain · 5 autonomy levels

Website · Quickstart · MCP Server · Tutorials

---

The Problem

MCP connects AI to software. But AI is entering the physical world — robots, drones, factories, farms, hospitals. There's no universal standard for how AI safely controls hardware.

Halyn connects AI to everything else. Servers, sensors, robots, industrial machines, smart homes — with safety rules the AI cannot break.

Quickstart

pip install halyn

from halyn import ControlPlane, SSHDriver

cp = ControlPlane()
cp.connect(SSHDriver("192.168.1.10", "admin"))

# Safety rules — the AI physically cannot bypass these
cp.shield("deny * delete *")
cp.shield("deny * write /etc/*")

# Observe
state = cp.observe()
print(state)  # {"cpu": 23.4, "mem": 67.2, ...}

# Act (within shield rules)
cp.act("restart nginx")     # ✓ allowed
cp.act("rm -rf /etc")       # ✗ blocked by shield

MCP Server

Connect Claude.ai, ChatGPT, Cursor, or any MCP client directly to your devices:

# Start the MCP Server
halyn-mcp --port 8935

# In Claude.ai → Settings → MCP → Add Server
# URL: http://your-server:8935/mcp

Then just talk:

• "Show me the status of all connected devices"
• "What's the temperature in the greenhouse?"
• "Restart nginx on server-01"
• "Show me the audit trail"

The AI sees your devices as tools. Shield rules still apply. Every action is audited.

12 Drivers

Driver	Target	Protocol
SSHDriver	Linux/Mac servers	SSH
HTTPDriver	REST APIs, webhooks	HTTP
WebSocketDriver	Real-time streams	WebSocket
SerialDriver	Arduino, Pi, PLCs	Serial
MQTTDriver	IoT sensors	MQTT
OPCUADriver	Industrial machines	OPC-UA
ROS2Driver	Robots, drones	ROS2/DDS
DDSDriver	Real-time systems	DDS
DockerDriver	Containers	Docker API
BrowserDriver	Web automation	Chrome CDP
UnitreeDriver	Unitree robots	UDP
SocketDriver	Raw TCP/UDP	Socket

Safety

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

cp.shield("deny * delete *")                    # no deletions
cp.shield("deny * write temperature > 180")     # thermal limit
cp.shield("deny lock unlock between 23:00-06:00")  # doors locked at night

Audit Chain — Every action recorded in SHA-256 hash chain. Tamper-evident. Append-only.

Consent Protocol — Devices and AI negotiate permissions before any interaction.

5 Autonomy Levels — From fully manual to fully autonomous. You choose.

Level	Name	Who Decides
0	Manual	Human does everything
1	Advisory	AI suggests, human acts
2	Supervised	AI acts, human approves
3	Monitored	AI acts freely, human watches
4	Autonomous	AI decides within shield rules

Works with Every LLM

# Claude
import anthropic
client = anthropic.Anthropic()
response = client.messages.create(model="claude-sonnet-4-6", ...)
cp.act(response.content[0].text)

# GPT
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(model="gpt-5", ...)

# Local (Ollama — no internet needed)
import ollama
response = ollama.chat(model="llama4", ...)

# All use the same shield rules. Same audit chain.

Built on NRP

Halyn implements NRP (Node Reach Protocol) — the open standard for AI-to-device communication. 6 primitives: Manifest, Observe, Act, Shield, Audit, Consent.

Free. Open Source. Forever.

MIT licensed. Use it for anything. Modify it. Build a company on it. No fees, no limits, no lock-in.

---

halyn.dev · nrprotocol.dev · PyPI · [e-mail]