AI Agent Architectures
I dug into the source code of 8 open-source AI agent frameworks to understand how they actually work. This is what I found -- and what you can steal for your own agents.
Each framework gets 30+ pages of annotated source code and architecture diagrams -- covering event loops, memory, tools, security, and LLM integration. The Comparison page puts all eight side by side. The Build Your Own guide turns the patterns into concrete advice for building your own.
Who is this for?
Software engineers who want to go beyond “call the API” and understand what's actually happening inside these systems -- event loops, memory architectures, tool sandboxing, context management, and security models.
How to use this site
- Start with the Comparison for a bird's-eye view
- Pick a framework page to read annotated source code
- Check Architecture Patterns for design themes that show up across frameworks
- Use Build Your Own when you're ready to start coding
The Frameworks
OpenClaw
TypeScriptMulti-channel personal AI gateway
IronClaw
RustSecurity-first Rust agent framework
PicoClaw
GoUltra-lightweight embedded agent
HermitClaw
PythonAutonomous research tamagotchi
Spacebot
RustMulti-agent delegation framework
pi (pi.dev)
TypeScriptMinimalist coding agent CLI
Hermes Agent
PythonPersonal autonomous agent with persistent memory
OpenAI Agents SDK
Python / TypeScriptMulti-agent workflow framework