Agent OS Documentation
May 8, 2026 · View on GitHub
Welcome to the Agent OS documentation. Agent OS is a kernel architecture for governing autonomous AI agents with deterministic policy enforcement.
Quick Navigation
🚀 Getting Started
| Guide | Time | Description |
|---|---|---|
| 5-Minute Quickstart | 5 min | Minimal setup, maximum speed |
| 30-Minute Deep Dive | 30 min | Comprehensive walkthrough |
| First Governed Agent | 15 min | Build a complete agent |
| Cheatsheet | - | Quick reference card |
📓 Interactive Notebooks
Learn by doing with our Jupyter notebooks:
| Notebook | Time | Description |
|---|---|---|
| Hello Agent OS | 5 min | Your first governed agent |
| Episodic Memory | 15 min | Persistent agent memory |
| Time-Travel Debugging | 20 min | Replay agent decisions |
| Verification | 15 min | Detect hallucinations |
| Multi-Agent Coordination | 20 min | Agent trust protocols |
| Policy Engine | 15 min | Deep dive into policies |
📚 Tutorials
- Using Message Bus Adapters - Connect agents with Redis, Kafka, NATS
- Creating Custom Tools - Build safe tools for agents
- Intent-Based Authorization - Declare, approve, and verify agent actions with drift detection
🏗️ Architecture
- Kernel Internals - How the kernel works
🔧 Reference
- Framework Integrations - LangChain, OpenAI, CrewAI
- AgentConfig File Reference - Supported YAML keys and validation rules
- Dependencies - Package dependencies
- Security Specification - Security model
- FAQ - Common questions and answers
📋 RFCs
- RFC-003: Agent Signals - POSIX-style signals
- RFC-004: Agent Primitives - Core primitives
🎯 Case Studies
- Carbon Auditor - Fraud detection example
- DeFi Sentinel - Attack detection
- Grid Balancing - Multi-agent coordination
Installation
# Core package
pip install agent-os-kernel
# With all features
pip install agent-os-kernel[full]
One-Command Quickstart
macOS/Linux:
curl -sSL https://get.agent-os.dev | bash
Windows (PowerShell):
iwr -useb https://get.agent-os.dev/win | iex
Hello World
from agent_os import KernelSpace
kernel = KernelSpace(policy="strict")
@kernel.register
async def my_agent(task: str):
return f"Processed: {task}"
# Run with kernel governance
result = await kernel.execute(my_agent, "analyze data")
Key Concepts
Kernel vs User Space
┌─────────────────────────────────────────────────────────┐
│ USER SPACE (Agent Code) │
│ Your agent code runs here. Can crash, hallucinate. │
├─────────────────────────────────────────────────────────┤
│ KERNEL SPACE (Agent OS) │
│ Policy Engine checks every action before execution │
│ If policy violated → SIGKILL (non-catchable) │
└─────────────────────────────────────────────────────────┘
Signals
Agent OS uses POSIX-style signals for control:
| Signal | Description |
|---|---|
SIGKILL | Terminate immediately (cannot be caught) |
SIGSTOP | Pause for human review |
SIGCONT | Resume execution |
Policies
Policies define what agents can and cannot do:
policies:
- name: read_only
deny:
- action: file_write
- action: database_write
IDE Extensions
| IDE | Status | Link |
|---|---|---|
| VS Code | ✅ Available | Marketplace |
| JetBrains | ✅ Available | Plugin |
| Cursor | ✅ Available | Extension |
| GitHub Copilot | ✅ Available | Extension |
Policy Templates
Pre-built templates for common use cases:
| Template | Use Case |
|---|---|
| secure-coding | General development |
| data-protection | PII handling |
| enterprise | Production deployments |
# Use a template
agentos init my-project --template secure-coding