Demo: AI agent discovering a TECNO phone, reserving it, executing an ADB command, and getting results back.
- AI agents can call APIs and write code, but they can't reliably control physical hardware.
- Every hardware integration today is bespoke, brittle, and non-composable.
- OAHL gives agents a standard lifecycle: discover → reserve → execute → release.
- One protocol for Android phones, cameras, SDR radios, sensors, robots, and anything with a transport plugin.
You don't need to write any code to connect common hardware.
# 1. Install globally
npm install -g @fredabila/oahl
# 2. Run the interactive setup
oahl init
# 3. Start the node
oahl start- Hardware Abstraction: Expose any physical device as standardized Capabilities to AI Agents.
- Access Policies: Declarative policies to strictly regulate which agent can use which device.
- Session Exclusivity: Ensure multiple AI agents don't conflict using device reservations.
- Transport Plugins: Native support for Serial (UART/USB), ADB, MQTT, TCP/IP, and more.
- See the API Reference and Security Guide.
- Visit registry.oahl.dev — discover and share hardware capabilities, adapters, and transport plugins.
- Read the full technical paper: Open Agent Hardware Layer: A Protocol for AI-Hardware Interoperability
graph TD
classDef highlight fill:#e3b341,stroke:#333,stroke-width:2px,color:#000;
classDef core fill:#2d3748,stroke:#4a5568,stroke-width:2px,color:#fff;
classDef hardware fill:#2f855a,stroke:#276749,stroke-width:2px,color:#fff;
subgraph "Agent Layer"
A[AI Agent / LLM]:::highlight -- "@oahl/sdk" --> B{Cloud Registry / Relay}
A -. "Local LAN (Optional)" .-> C
end
subgraph "Hardware Node Environment"
B -- "Secure WebSocket" --> C[OAHL Node Server]:::core
C --> D[Policy & Auth Engine]:::core
D --> E[Session Manager]:::core
E --> F[Hardware Adapters]:::core
F -- "Transport Plugins<br/>(Serial, BLE, ADB)" --> G((Physical Devices)):::hardware
end
- See CONTRIBUTING.md for step-by-step setup instructions, development workflow, and adaptation guides.
- Ready to help? Check out our list of "good first issues" to get started!
See the docs/ folder for:
- Architecture Overview
- Hardware Owner Guide
- Adapter Guide
- Agent Integration Guide
- W3C WoT Alignment 75ED
The @oahl/cloud package contains the hosted infrastructure that connects agents and nodes. When your node starts up, it securely registers itself with the Cloud Registry. AI Agents connect to this Cloud Registry (not your node directly) to request access to capabilities.
Why should I use Docker instead of running via npm start?
Docker is highly recommended because it avoids host machine pollution. Many hardware adapters (especially those communicating over serial or BLE) require native C++ bindings to be compiled during installation (e.g., serialport).
If you use npm start, you may need to install heavy dependencies like Visual Studio C++ Build Tools or python/make on your host machine. By using Docker, the node runs in a pre-configured Linux container with all build tools already installed.
Can I run OAHL completely offline / locally without the Cloud Registry?
Yes, absolutely. The Cloud Registry is strictly optional and used for global discovery. To run a 100% private, local-only node, simply open your oahl-config.json file and remove or empty the cloud_url property.
Is it secure to let an AI control my hardware?
OAHL is designed specifically to make this safe:
- Container Sandboxing: Hardware runs in isolated Docker containers; the agent cannot access the host machine's filesystem.
- Access Policies: You write a declarative policy for every device outlining exactly which capabilities an agent is allowed to execute (and you can ban specific agents/organizations).
- Session Exclusivity: An agent must request a "session" to use a device. While an agent holds a session, no other agent can interrupt or send contradictory commands to your hardware.
Copyright © 2024–2026 Fred Abila and the OAHL Contributors.