Coming soon · 2026

Scattered memory, tidied into one.

A private memory index shared by all your AI agents.

Get notified View on GitHub

Scroll to tidy up

Your memory is all there. Just in too many places.

Scattered memory

Meeting recordings sit in Plaud, plans in Notion, a contract in some folder on your laptop, photos on your phone. When you need something, you first have to remember where you put it.

Your AI agents are no better off. Claude Code remembers per folder, the phone app keeps its own notes, and the browser agent keeps yet another set. None of them know what the others know, so you explain the same things again and again.

Wherever it lives. Whatever it is. Whoever asks.

Features

  • Across your devices

    A lightweight agent indexes quietly on your computers. Your iPhone searches the same index.

    • Mac
    • Windows
    • iPhone

  • In any format

    Documents, photos and videos, plus the cloud apps you work in.

    • PDF
    • Office
    • HWP
    • Photos
    • Videos
    • Notion
    • Google Drive
    • Gmail
    • Plaud
    • Slack

  • Korean-first search

    KURE-v1 embeddings and Kiwi morphological search find things by meaning, not just matching words. Image vectors make what's in your photos searchable, too.

  • Any agent, via MCP

    Claude Code, Codex, claude.ai on the web and iOS, and browser agents all search through the same MCP endpoint.

Three steps. Originals stay put.

How it works

  1. Index on your device

    Files are read, parsed and turned into vectors right where they live.

  2. Send only excerpts and vectors

    They go to a small server you run yourself, a $24/month VPS. Originals never do.

  3. Search from any agent

    Ask from Claude Code at your desk or claude.ai on your phone. The answer comes from the same index.

Your originals never leave your devices.

Privacy

  • Originals stay on device

    Your server keeps only the short excerpts and vectors that search needs.

  • Secrets never leave

    Files holding passwords, API keys or private keys never leave your device, not even as excerpts.

  • Self-hosted, open source

    You run the server, and the code is MIT-licensed, so you can check exactly what goes where.

Works with

tidy-one is an independent open-source project and is not affiliated with or endorsed by any of the services above. Product names and logos belong to their owners.

Where things stand.

Roadmap

Done

  • Shared memory across devices
  • Semantic search
  • Push-to-deploy server

Next

  • Document & photo indexing
  • Video indexing
  • Cloud sources
  • Windows indexing agent

First release in 2026. The order may change.

Tidy the pixels

00/12

A small game: click or tap the scattered pixels to put them back in the grid