Verdict
Nanobot on a Pi 5 (8 GB) is the most quietly capable setup at this price tier. The Python codebase fits trivially in the available RAM, and you can pair it with Ollama running Phi-3 mini for an entirely local, entirely auditable, entirely cheap agent stack. No sandbox by design, single-user by design — match those to your threat model first.
Setup notes
Pi OS Bookworm, Python 3.11, virtualenv. Install Nanobot from PyPI. Install Ollama for ARM (the Pi 5 builds work). Pull Phi-3 mini 3.8B Q4. Point Nanobot at Ollama.
Performance
Nanobot itself: ~80 MB RAM, ~3% CPU idle. Phi-3 mini Q4: 6 tok/s. Total stack: comfortably under 4 GB RAM, leaves headroom for tools and OS.
What breaks
- Multi-user
- Anything requiring 7B+ local model capability
Want to know more
See the full Nanobot review and the Raspberry Pi 5 buyer's notes.