DEV Community

Nikita Dankov
Nikita Dankov

Posted on

I built a privacy-first AI search that runs completely locally

I built a privacy-first AI search that runs completely locally

Modern computers contain thousands of files: code, photos, documents, screenshots.
But finding the right file is still surprisingly difficult.

Most search tools rely on keywords.

So if you don’t remember the exact file name or text inside it, you're often stuck.

I wanted something different.

Something that works more like human memory.

So I built an open-source tool called Smart Local Search.

GitHub:
https://github.com/dan99nik/smart-locale-search

The problem

Traditional file search works like this:

You type a keyword:

  • invoice
  • report
  • photo_2023

But what if you want to search like this?

photo with mountains and a lake
jwt authentication middleware
diagram of system architecture

Standard search engines can't understand meaning.

They only match text.

The idea

I wanted to build a semantic local search engine.

Something that allows you to search your computer the way you think.

Example queries:

  • photo with mountains at sunset
  • python jwt authentication
  • database schema diagram

Instead of keywords, it understands descriptions.

Key principles

The project follows three simple principles.

1️⃣ Everything runs locally

No cloud.
No external APIs.
No data leaves your computer.

2️⃣ Privacy first

Your files stay on your machine.
No telemetry.
No analytics.

3️⃣ Open source

You can inspect the code, modify it, and build on top of it.

GitHub:
https://github.com/dan99nik/smart-locale-search

What it can search

Currently the project can search:

  • files
  • photos
  • code

Using natural language queries.

Example:

  • photo of mountains and lake
  • authentication middleware
  • api routing logic

Why I built it

I realized something:

Our computers contain massive knowledge bases.

But we still search them like it’s 2005.

Keyword search doesn't scale with modern workflows.

Especially when working with:

  • large codebases
  • photo libraries
  • research folders
  • documentation

Semantic search makes this much easier.

Demo idea

Example workflow:

Search query:

  • jwt authentication middleware

Result:

  • src/auth/jwt_middleware.py

Or:

Search query:

  • photo with mountains and lake

Result:

  • IMG_2023_08_mountains.jpg

Future plans

Some ideas for the roadmap:

  • VSCode extension
  • better indexing
  • faster search
  • plugin system
  • UI

Looking for feedback

This is an early version and I’d love feedback from developers.

Questions I'm curious about:

  • Would you use semantic local search?
  • What would you search for?
  • What features would be useful?

GitHub:
https://github.com/dan99nik/smart-locale-search

Final thought

We spend years building knowledge on our computers.

Maybe it’s time we had better ways to search it.

Top comments (0)