Search and AI is where you decide how much work Relic does on your items after it saves them. All of it runs on your computer. Nothing on this pane sends anything anywhere. Every row below is drawn the way the app draws it, with its default.
On-device smart tags
This is the master switch. Turn it off and every row under it disappears and no analysis happens at all. Search still works on the words in your items.
Under the row is a short status line in small type, and it is the thing to read when you are wondering why an item has no tags yet.
- Off
- The switch is off. Nothing is being analysed.
- Downloading models (~750 MB, one-time)…
- First run. It happens once, in the background, and you can keep using Relic.
- Basic tags now · loading ML models…
- Simple tags are already being written while the models finish loading.
- Ready: semantic search off (embeddings disabled)
- Everything works except search by meaning, because you turned that row off.
- Ready: full local ML
- All of it is on and running.
- Engine not found
- The part of Relic that does this work is missing from the install. Reinstall from the download page.
When there is a queue, a second line appears under the status, for example tagging 1,204 items…. That is normal after you turn something on or save a lot of items at once. It works through them in the background.
What each part does
These three rows only appear while the master switch is on. All three start on.
- Image text (OCR) reads the text in screenshots and photos so you can search them. This is the one that makes a screenshot of a receipt findable by typing the shop name.
- Smart content tags tags what a picture shows, like a person, an animal or a product. With it off, an image still gets its main category.
- Semantic searchlets you search by meaning, so “the thing about the flight” can find a boarding pass that never uses the word flight. Off, search only matches the words that are actually there.
Describing items
Describe items writes a short title and a few topic tags for the things you keep, so a photo with no text in it is still findable and a note gets a name without you typing one. It is off to begin with because it downloads another set of models and takes about a second per item. Turn it on if your vault has a lot of photos.
Describe history too only appears once the row above it is on. It widens the same work to clipboard items you never kept. Most copies are throwaway, so this stays off unless you really want everything covered.
Speed and the downloaded models
Analysis speed is how much of your processor the background work may take. Gentle is for an old or busy machine and keeps everything else responsive. Fast gets through a backlog sooner and competes with whatever you are doing. Balanced sits between the two and is the default.
Downloaded ML models shows how much disk the models take, with a Remove button to get it back. The row is hidden until something has actually been downloaded. You have to turn the master switch off first, and the row says so. Removing them does not lose your tags. If you turn smart tags back on later they download again.
Personalized ranking
On by default. Relic watches which rows you pick, both from a search and when you paste, and floats those higher next time. It is learned on this computer and stays there. It never syncs and it never leaves.
The second row only appears while ranking is on. Clear learned ranking forgets all of it in one press and the link reads Cleared for a moment. Use it if the order in the window has settled around items you no longer care about. Your items are untouched.