I built myself a notebook with AI search — it runs 100% locally
A normal planner was hard to search, so I built my own notebook with AI search. It runs locally on a Raspberry Pi and my computer - no note goes to the cloud.
I could have kept my notes in a normal paper planner, but there was no sensible way to search them. So I built my own - with AI search.
Someone might ask: “Hold on - you added an AI search engine to your private notes? Now everyone will know about them, from Apple to OpenAI, maybe Anthropic too”.
Nothing could be further from the truth. Everything runs locally, on my own hardware, through the Ollama engine. No note leaves my network.
What it actually does
The app stores notes and tasks, which I view as a Kanban board - active, waiting, someday, done. Thanks to the AI search I can ask it in plain language: “what do I have planned for next week” or “how much do I have to do on Friday”, and I get an answer based only on my own records - no hallucinations. The model does not guess anything, it only searches what I actually wrote down.
How I built it
I wrote down the assumptions, described how everything should work, designed the interface and sat down with Claude Code - with the full context right from the start.
Three models work in the project, each for a different job:
- snowflake-arctic-embed2 — semantic search, that is understanding the meaning of a question, not just matching words.
- Bielik Minitron 7B (previously qwen3:8b) — answers the questions. In my view it is faster than its predecessor and better at pulling data out of Polish text.
- gemma3:4b — recognises images. It kicks in when I upload a photo and writes a description for it.
The end result
The app itself - notes, tasks, the Kanban board - runs at my place around the clock on a Raspberry Pi with an encrypted card, and is available on all my devices through the private Tailscale network. The AI engine is a separate story: because of the resources it needs, I only start it when I turn on my computer, so I use it when I really need it.
My small project has grown to the point where it became a great testing ground for working with AI. And the deeper into the woods I go, the more trees there are… and ideas. 😉