Verdict
We would not host Nanobot on a Pi Zero 2 W. Nanobot needs Python 3.11 or newer and ships Linux ARM64 wheels, so the board, whose quad-core Cortex-A53 is 64-bit, would have to run a 64-bit OS, and nothing says its 512 MB is enough: Nanobot's docs state no minimum memory and we have not measured it. Until October 2026 this page suggested running Nanobot on the Zero as an MCP server for a bigger agent elsewhere; Nanobot's docs describe it as an MCP client that connects to MCP servers, and document no way to run it as one.
Setup notes
If you try anyway: a 64-bit OS, Python 3.11+, pip install nanobot-ai in a virtual environment, then nanobot onboard. The model runs at your provider, not on the board. Nanobot's own server mode, nanobot serve, exposes an OpenAI-compatible API (127.0.0.1:8900 by default), not MCP tools. To give an agent tools that live on the Zero, run a dedicated MCP server there and connect Nanobot, or another agent, to it as a client.
Performance
No measurement of ours. The board has 512 MB of LPDDR2 (Raspberry Pi product brief), shared by the OS, Python and the Nanobot process.
What breaks
- Running Nanobot as an MCP server: its docs only describe an MCP client
- A local model on the board
Sources
The agent's defaults and settings above were read in its own documentation on 2026-10-18.
- Nanobot README (v0.3.5): install, platform wheels
- Nanobot docs (v0.3.5): providers, including Ollama
- Nanobot docs (v0.3.5): configuration, security section
- Nanobot SECURITY.md (v0.3.5): exec sandbox
- Nanobot docs (v0.3.5): configure MCP tools (Nanobot as an MCP client)
- Nanobot docs (v0.3.5): OpenAI-compatible API (nanobot serve)
- Raspberry Pi Zero 2 W product brief: 64-bit Cortex-A53, 512 MB LPDDR2
Want to know more
See the Nanobot page and the Raspberry Pi Zero 2 W spec sheet.