A deep technical comparison of 7 open-source AI agent frameworks, built for software engineers who want to understand how AI agents work under the hood — and how to build their own.
| Framework | Language | Focus |
|---|---|---|
| OpenClaw | TypeScript | Multi-channel personal AI gateway |
| IronClaw | Rust | Security-first agent framework |
| PicoClaw | Go | Ultra-lightweight embedded agent |
| HermitClaw | Python | Autonomous research tamagotchi |
| Spacebot | Rust | Multi-agent delegation framework |
| pi (pi.dev) | TypeScript | Minimalist coding agent CLI |
| Hermes Agent | Python | Personal autonomous agent with persistent memory |
Each analysis covers architecture, memory systems, tool calling, LLM integration, security, multi-channel support, state management, and more — with annotated code snippets and Mermaid diagrams throughout.
- Individual framework analyses — 30+ pages each of annotated source code and architecture breakdowns
- Cross-framework comparison — synthesized findings across all seven frameworks
- Architecture patterns — recurring design themes and trade-offs
- Build your own guide — actionable patterns distilled into a practical reference
├── analyses/ # Markdown source for each framework analysis
├── site/ # Next.js site that renders the analyses
│ └── src/
│ ├── app/ # Pages (home, comparison, patterns, build-your-own, framework/[slug])
│ ├── components/
│ └── lib/ # Markdown parsing and framework metadata
├── analysis-prompt.md # Template used to generate each analysis
└── package.json # Root package.json (proxies to site/)
# Install dependencies
npm install
# Run the dev server
npm run dev
# Build for production
npm run buildThe site runs at http://localhost:3000.
- Write a markdown file in
analyses/following the template inanalysis-prompt.md - Add the framework metadata to
site/src/lib/frameworks.ts - The site will automatically pick up the new analysis at
/framework/<slug>
Want to see a framework or agent harness added? Open an issue with a link to the project and a brief description of why it's interesting. See CONTRIBUTING.md for more details.
MIT