Your mind for what you find.
Save inspiration from anywhere. Find it again by describing it.
⬇ Download for Windows · Tour · Why lompii · Get started · FAQ
Type what you remember, not what you named it. Watch in full quality.
You saved it months ago: a dark website with huge type, a poster in a particular orange, a line someone wrote that stuck. You don't remember where it was, what it was called, or which folder it went into.
lompii is a desktop app that remembers for you. Save images, page captures, quotes and snippets from your browser with one right-click. lompii looks at each one and works out what it shows and what it's about. Later, type what you remember, "grainy portrait, deep shadow", and it comes back.
It's made for designers, illustrators, photographers, writers, and anyone who collects references.
| Right-click an image or text on any site, capture part of a page, or drop in a file. | Each save gets a title, a summary, tags and its colours. You file nothing. | Describe it in your own words. lompii matches how things look and what they mean. |
Most places you keep inspiration are good at storing it and bad at giving it back.
| Folders on your computer | Browser bookmarks | Cloud inspiration apps | lompii | |
|---|---|---|---|---|
| Find something by describing it | ✗ | ✗ | some | ✓ |
| Find images by how they look, not just their words | ✗ | ✗ | some | ✓ |
| Keeps the image itself, even if the page disappears | ✓ | ✗ | ✓ | ✓ |
| Save from any website in one click | ✗ | ✓ | ✓ | ✓ |
| Your library stays on your computer | ✓ | ✓ | ✗ | ✓ |
| Works without an account | ✓ | ✓ | ✗ | ✓ |
| Free | ✓ | ✓ | usually paid | ✓ |
| Choose your own AI, or run one offline | n/a | n/a | rarely | ✓ |
What that means in practice:
- Finding beats filing. No folders to keep tidy. Search by look ("mountain landscape"), by meaning ("an AI meeting assistant"), or by the words written inside an image.
- Private by design. No account, no lompii server. Search runs on your own computer, and nothing leaves it except the item you send to the AI provider you chose. Choose a local model and nothing leaves at all.
- Free, with any AI. Bring a key from Anthropic, OpenAI, Google, OpenRouter, or run Ollama or LM Studio. Search by description works with no AI key at all.
The browser extension adds Add to lompii to the right-click menu on every site. Save an image
at the largest size the site offers, a quote or snippet of selected text, or a capture of the
visible area, the full page, or an area you drag over (Alt+Shift+S). Each save keeps the
address it came from. You can also drag image files onto the window, or paste one with Ctrl+V.
Type a few words about what you remember. lompii's local models compare your words with what each
image looks like (CLIP) and what each save means (MiniLM), alongside titles, tags, notes and
the text inside images. Press Enter to keep a word as a chip and narrow things down.
Open any card to see it large, with its colour palette (click a swatch to copy the hex), a short AI summary, tags, the source link and your notes. Generate prompt writes a text-to-image prompt (Midjourney, DALL·E, Stable Diffusion) that would recreate the image.
Found one thing you like? Same vibe lines up everything in your library that's closest to it, by look, meaning, tags and colour.
Stacks are collections you build by hand: a poster wall, a moodboard for a client, a set of type references. Give each one a colour, add cards from anywhere, or turn any search into a stack in one click.
Where is it? Click the lompii logo at the top left. That's the archive's door.
- To archive a card: drag it onto the logo, or select it and press
A, or click the archive button in its detail view. It leaves Everything and Stacks but is kept, with its own search. - To bring it back: open the archive and drag the card onto the logo again (or press
A). - To lock it: set a 6-digit PIN in Settings → General. The archive asks for it each time it opens.
The grid in light or dark mode (this image follows your GitHub theme). Three card sizes.
- Trash with undo. Deleted items stay for 30 days and can be restored.
- Select several cards (
Ctrl/Shift-click) to stack, tag, archive or delete them together. - Tags of your own beside the AI's, and rename, merge or delete them across the library.
- Runs in the tray so the extension can always save, and
Ctrl+Shift+Lbrings it forward from any app. - One-click backup of the whole library to a folder you choose.
Download lompii Setup (.exe) from the
latest release and run it.
The installer isn't code-signed yet, so Windows SmartScreen may warn you the first time. Choose More info → Run anyway.
Settings → Extension in lompii walks you through it.
Chrome, Edge, Brave, Opera, Vivaldi or Arc
The extension comes with the app.
- In lompii, open Settings → Extension and click Show folder.
- Open
chrome://extensions(oredge://extensions,brave://extensions…), turn on Developer mode, click Load unpacked and choose that folder.
Firefox, Zen, LibreWolf or Floorp
The extension is signed by Mozilla.
- Open the install link in your browser.
- Open the download and click Add.
Then click Connect on the page that opens. lompii comes forward showing a four-digit code; if it matches the one in the browser, click Allow. That's it: right-click any image and choose Add to lompii.
| How | What it saves |
|---|---|
| Right-click an image → Add to lompii | the image, at the largest size the site offers |
| Select text, right-click → Add to lompii | a quote or a snippet |
| Right-click the page → Capture an area… | part of the page |
| Toolbar button | the visible area, full page, a selected area, a delayed capture, or selected text |
Alt+Shift+S |
a selected area (Enter takes the whole view, Esc cancels) |
Search by description and Same vibe work without any key. An AI provider adds titles, summaries and tags. Open Settings → AI providers and choose one:
| Provider | You need | Default model |
|---|---|---|
| Anthropic (Claude) | an API key | claude-opus-5 |
| OpenAI | an API key | gpt-4o-mini |
| Google Gemini | an API key | gemini-3.8-flash |
| OpenRouter | an API key | google/gemini-3.8-flash |
| Ollama | Ollama running on your computer | any model you have |
| OpenAI-compatible server (LM Studio, LiteLLM, vLLM…) | its address, and a key if it asks | any model it serves |
Any model that reads images will do. You pay your provider directly for what you use, or nothing with a local model. New saves are analysed automatically; Analyse now catches up on older ones.
Is it really free?
Yes. lompii is MIT-licensed open source with no paid tier. The only cost is whatever your AI provider charges for analysing your saves, and that's zero with a local model like Ollama.
Does anything leave my computer?
Your library, its database and its search never leave. When an item is analysed, that one item (a resized image or its text) goes to the AI provider you chose. The search models download once from Hugging Face, and the app checks GitHub for updates. That's all. See Privacy.
Do I need an AI key?
No. Search by description, Same vibe, stacks, tags and the archive all work without one. A key (or a local model) adds automatic titles, summaries and tags, which make search even better.
Where's the archive?
Click the lompii logo at the top left. Drag cards onto the logo to archive them, and drag them onto it again from inside the archive to bring them back. See The archive.
Is there a Mac or Linux version?
Not yet. lompii is Windows-only for now; a macOS build is on the roadmap.
Keyboard shortcuts
| Key | Action |
|---|---|
Ctrl+Shift+L |
Bring lompii forward, from any app |
Ctrl+K |
Search |
Enter in search |
Keep the typed word as a chip |
Backspace in an empty search |
Remove the last chip, then the filter |
| Arrow keys | Move between cards |
Enter |
Open the selected card |
A |
Archive the selected cards (unarchive, in the archive) |
Delete |
Move the selected card to the trash |
Ctrl-click / Shift-click |
Select several cards / a run of cards |
Esc |
Close, clear or step back |
Ctrl+V |
Save a pasted image |
You can change the extension's shortcut at chrome://extensions/shortcuts, or in Firefox from
about:addons → ⚙️ → Manage Extension Shortcuts.
lompii has no account and no server. Everything lives in %APPDATA%\lompii\ on your computer:
| Path | What's in it |
|---|---|
lompii.db |
the library: items, tags, stacks and search data (SQLite) |
library/, thumbs/ |
your files and their thumbnails |
models/ |
the search models, downloaded once on first run (about 170 MB) |
settings.json |
your preferences; AI keys are encrypted with Windows' secure storage |
- Your library stays on your computer. When an item is analysed, or you ask for a prompt, that item (a resized image or its text) goes to the AI provider you chose, with the site's name, the page title and some of your existing tags so it can reuse them. With Ollama, or a server on your own computer, nothing leaves at all.
- Two other connections: the search models download once from Hugging Face, and the installed app checks GitHub for new versions. Neither sends anything from your library.
- The extension talks to the app over
127.0.0.1only, with a key handed over when you approve the four-digit code. Web pages can't reach your library. - The archive PIN is stored only as a salted hash, and five wrong guesses in a row slow further tries. It keeps the archive private inside lompii. It doesn't encrypt the files on disk.
The full policy is in PRIVACY.md.
For developers: run, test and package lompii
You'll need Node.js 22 or newer and Windows.
git clone https://github.com/lompii-app/lompii.git
cd lompii
npm install # also builds the native modules for Electron
npm run dev # runs the app, reloading as you edit| Script | What it does |
|---|---|
npm run dev |
Run in development with hot reload |
npm run typecheck |
Type-check the project |
npm test |
Run the tests |
npm run build |
Type-check and build to out/ |
npm run dist |
Build the Windows installer into release/ |
Set LOMPII_DATA=<folder> to run on a separate, throwaway library.
After changing anything in extension/, click the reload icon on lompii's card in
chrome://extensions. To try it in Firefox, use Load Temporary Add-on… in
about:debugging#/runtime/this-firefox; a permanent install needs Mozilla to sign it.
Browser extension ─ HTTP, 127.0.0.1 ──┐
▼
┌──────────── Electron main process ────────────┐
│ server.ts local API for the extension │
│ ingest.ts dedupe, thumbnail, colour, SQLite │
│ ai.ts analysis queue and providers │
│ embed.ts local CLIP and MiniLM search │
│ index.ts IPC, window, lompii:// files │
└───────────────▲───────────────────┬───────────┘
IPC (preload) live updates
┌───────────────┴───────────────────▼───────────┐
│ Renderer: React + Tailwind │
└───────────────────────────────────────────────┘
| Folder | What's in it |
|---|---|
src/main/ |
the Electron main process: database, saving, AI, search, the extension's API |
src/preload/ |
the typed bridge between the main process and the window |
src/renderer/ |
the React app |
src/shared/ |
types and helpers both sides use |
extension/ |
the browser extension for Chrome and Firefox (Manifest V3, plain JavaScript) |
test/ |
tests for parsing, scoring and the PIN |
Issues and pull requests are welcome:
report a problem or suggest an idea.
A git hook runs the typecheck and tests before every push (npm install turns it on), so fix
what it reports rather than skipping it.
- Windows only for now. macOS and Linux builds haven't been tested yet.
- Full-page capture stitches screenshots as it scrolls. Pages that scroll an inner panel come out as one screen, and very long pages stop at about 20,000 px.
- One filter at a time (a kind, a stack or same vibe), plus any number of search words.
- Smart stacks that keep filling from a saved search
- Restoring a backup from inside the app
- macOS build and code signing
- Provider logos from Lobe Icons (MIT).
- Search runs CLIP and MiniLM through Transformers.js.
- Screenshots use a throwaway library of generated artwork.
MIT © 2026 Midhun






