Build an assistant that answers questions about your own notes

Here's an evening idea for anyone whose notes are scattered and impossible to search. Let's build a helper you feed your notes to — then ask in plain language, and it answers, pointing at the right note.

What it is and why

It's a little "search by meaning" over your own text. Not "find the exact word" but "find by sense": you ask "what did I plan to buy my parents?" and it gets it, even if the note says "gift for mom and dad".

Under the hood is a trick called RAG: the model first pulls a couple of relevant chunks from your notes, then answers from them, not from memory. So it doesn't make things up — it answers from your own text.

What you'll need

• your notes as a set of plain text files (export from your notes app to .txt/.md);
• a free model — via OpenRouter (no card) or local, now that local models got faster;
• an editor with an AI agent (Cursor, or any one where you can ask for code "in words").

How to approach it

Don't write the code by hand — describe the task to the agent. Give it this start:

Build a simple local Python CLI: it reads all .txt and .md files from a notes/ folder, splits them into chunks, builds a semantic search (embeddings + a local vector store like FAISS or chromadb). For my question it finds the 3 most relevant chunks and passes them to the model as context so it answers only from them and cites which note it used. Get the model key from an env var. Make it so I can run python ask.py "my question".

Then iterate: run it, look, ask the agent to fix. That's vibe coding.

What's in it for you

You'll build your first thing with RAG — the single most useful trick for "AI over my own data". You'll get it hands-on: chunks → search by meaning → answer from context. The same skeleton later becomes a docs bot, a study assistant, or search over your work files.

Small but genuinely useful: an evening's work, and your notes finally answer questions.