SwarmAI

The four ideas

SwarmAI is described as a self-evolving Agent OS where every interaction sharpens how the system judges, not just what it knows. It serves four ideas. Self-evolution turns recurring mistakes into code gates in security_hooks.py rather than logged lessons. Brain-first treats every project as a domain brain with a six section structure that grows and decays. The Agent OS keeps cognition between sessions via hooks. Proprioception lets the desktop app sense and drive its own UI, with TSCC showing the real cognition behind a turn.

Architecture and usage

SwarmAI separates cognition, the OS, from knowledge, the disk, and argues that one edited line in SOUL.md shifts judgment more than a thousand memory rows, with every change a git diff. A dozen plus live guards sit in security_hooks.py covering commit, pytest and dangerous command gates. The quick start clones the repo, runs uv sync in the backend and npm install plus tauri:dev in the desktop, and requires Node.js 18 or newer, Python 3.11 or newer, Rust, uv and the Claude Code CLI.

Posture

The project is built with the Claude Agent SDK and provides both English and Chinese READMEs. The desktop app is inspectable: TSCC (Thread Scoped Cognitive Context) shows which files loaded, the token budget, every recall hit and its score, the security scan and the full prompt. The README frames the value as compounding across sessions rather than starting from zero each time. It is licensed MIT.

Editorial conclusion

SwarmAI is MIT licensed and reported thirty nine stars at indexing, with a macOS Apple Silicon disk image available from releases and a bilingual README.