autonomous-ai/autonomous-os

What it solves

Autonomous OS is designed to move robots from scripted demos and remote-controlled operation to true autonomy. It provides a standardized software stack that allows robots to think, act, and learn from experience without requiring the user to write complex low-level code for every new device.

How it works

The system is built as a modular, layered architecture where each component is swappable:

• Agentic Runtime: An engine (such as Hermes or Claude Code) that handles reasoning and decision-making based on the robot's identity and installed skills.
• Skills: Modular behaviors defined in Markdown files. Skills act by sending hardware markers to the system, allowing the agent to trigger physical actions without touching hardware pins directly.
• HAL (Hardware Abstraction Layer): Manages the physical capabilities (audio, vision, motion, etc.) and ensures that only declared hardware is mounted. It includes a safety gate to enforce physical bounds like speed and brightness.
• System Services: A Go-based daemon (os-server) that handles intent, agent switching, and OTA updates.
• Configuration: Robots are defined by four Markdown files (ROBOT.md, SOUL.md, SAFETY.md, and SKILL.md) which describe the body, personality, and safety constraints.

Who it’s for

• Robot developers who want a plug-and-play operating system for their hardware.
• AI researchers looking to deploy agentic reasoning engines onto physical embodied intelligence.
• Hobbyists who can add autonomy to existing robots (like Reachy Mini) or build their own using supported boards (Raspberry Pi, OrangePi).

Highlights

• Swappable Brains: Easily switch between different LLM-based reasoning engines live via a web UI.
• Skill Store: Install pre-made behaviors or create new ones using natural language in an app.
• Hardware Agnostic: Supports multiple boards and robots through a simple declaration system rather than code forks.
• Learning Loop: Built-in ability to create and refine skills from experience and search past conversations.