Projects / Your documents

Search your own documents privately

Ask questions across your PDFs and notes and get pointed to the right page, without uploading anything anywhere.

At a glance

You'll get
A simple search over a folder of your files that understands meaning, not just exact words.
Time
About 45 minutes
Difficulty
Moderate
Computers
Basic laptop (8 GB), Regular laptop or desktop (16 GB), Gaming PC (NVIDIA graphics card), Newer Mac (32 GB or more)
Cost
Free
Under the hood
Local RAG with EmbeddingGemma + a local vector store. Models: EmbeddingGemma (or similar small embedding model)

Know the limits medium risk

  • It can miss relevant passages or surface the wrong ones.
  • If you pair it with a chat model, the written answer can misquote or misread your documents.
  • Always click through to the source file and read the original before relying on it.

More in Using local AI safely.

Check the results before you rely on them.

The easy way: have an AI assistant do itRecommended if you're new to this

Open an AI coding assistant on the computer you want to set up (Claude Code, Codex, and OpenCode all work), paste the instructions below, and follow along. It will check your computer, install what's needed, and confirm everything works.

What's an AI coding assistant?

It's a version of ChatGPT-style AI that can run commands on your computer for you, with your permission. You talk to it in plain English, and it does the typing.

Before you paste: this lets an AI assistant run commands on your computer. Read what it plans to do before you approve each step, and say no to anything you don't understand.

Set up private, fully local document search on this machine using a small embedding model (prefer EmbeddingGemma or another compact embedding model that fits in RAM).

Goals:
1. Ask me which folder(s) to index (default to a small sample folder if I have not said).
2. Detect available RAM/GPU and pick an embedding model that fits.
3. Install only local dependencies (no cloud embedding APIs).
4. Chunk, embed, and store vectors on disk.
5. Provide a simple CLI or script: `query "..."` that prints the top matching chunks with file paths.
6. Verify with a question whose answer appears in only one sample file.

Constraints:
- Do not upload document contents anywhere.
- Do not invent benchmark numbers.
- Keep the setup reproducible: write down exact model id, versions, and commands in a short README in the project folder.

When finished, tell me how to re-index and how to query.

Do it yourselfFor hands-on folks, or if you want to learn how it works

Before you start

Pick a folder of documents to try it on: manuals, notes, PDFs. Start small, a few dozen files is plenty. You'll need Python 3.10 or newer.

What does this mean?

This uses an embedding model: a small AI that turns each paragraph into a list of numbers describing its meaning. When you ask a question, it finds paragraphs with similar meaning. These models are tiny, so even an older 8 GB laptop can do it.

Do it yourself

  1. Install Python and create a project folder.
  2. Install a local embedding library and a small vector database (exact versions pinned after testing).
  3. Split your documents into short chunks of text.
  4. Turn each chunk into numbers with the embedding model and save them on disk.
  5. Ask a question: the script finds the closest chunks and shows the file and passage.
What is a "vector database"?

A file that stores those lists of numbers and can quickly find the closest matches. Chroma and sqlite-vec are free ones that live on your computer.

How to tell it worked

  • Ask about something mentioned in only one file. That file comes up first.
  • Nothing needs the internet after the first download.

Test results

Results pending. We'll publish speed and quality on each of our test computers once we've run it. Until then, no numbers.