Verdict
Nanobot on a 32 GB mini PC leaves plenty of room for a local model next to the agent. Pair it with Ollama, one of Nanobot's built-in providers, for local-LLM-backed tasks, and keep a cloud provider (Anthropic, OpenAI, OpenRouter) as a fallback if you want one. Switch on its bubblewrap sandbox before it touches a chat app: it is off by default. Until October 2026 this page called Nanobot 4,000 lines to read in an afternoon, with no sandbox; at v0.3.5 it is about 160,000 lines of Python by our count, and the sandbox is opt-in.
Setup notes
Python 3.11+ and pip install nanobot-ai in a virtual environment (or uv tool install nanobot-ai). Add Ollama with Mistral 7B Q4 and point Nanobot's ollama provider at it. Install bubblewrap (apt install bubblewrap), then set tools.exec.sandbox to "bwrap" and tools.restrictToWorkspace to true, as its docs ask for production.
Performance
The constraint is the LLM endpoint, not Nanobot. No published llama.cpp figure for this chip's graphics; on the Vulkan scoreboard, Intel's Core Ultra 100 series writes 8.21 tokens per second with a 7B model at Q4_0.
What breaks
- A shared agent with per-user accounts (Nanobot is built as a personal agent)
- Network isolation: its bubblewrap sandbox confines files, not network access
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
Want to know more
See the Nanobot page and the Intel NUC 13 and Geekom IT13 spec sheet.