April 5, 2026
Local or LOL-cal?
Recall – local multimodal semantic search for your files
Find photos with words—and ignite a privacy brawl
TLDR: A new tool called Recall lets you find photos and files by typing simple phrases, powered by Google’s Gemini and a local database. The community is split between wowed convenience and worries: privacy concerns over data leaving the computer, doubts about usefulness for code, and spicy trust drama around the project’s repo.
“Recall” promises a magic trick: type “team dinner” and it digs up the photo—even if the pic has zero labels. It builds a local “memory” of your files and lets you search by plain English across images, audio, video, PDFs, and text. There’s even an animated setup wizard and a slick Raycast visual search. The catch that lit up the comments: it uses Google’s Gemini to create those smart file “fingerprints,” so some requests leave your machine.
Supporters are hyped. One fan sums it up as local search with cross‑media smarts, insisting “nothing leaves your machine except embedding API calls.” But the peanut gallery arrived with pitchforks and memes. Privacy hawks call “local” a marketing stretch—“Everything you have on your PC is sent to AI… ‘local,’ yeah,” sneers one commenter, dropping a link to old‑school Recoll for text‑only search. Meanwhile, a spicy subplot: a user side‑eyes the project’s fresh GitHub activity and mentions “commits coauthored by Claude,” throwing around the words supply chain like it’s a horror movie.
Pragmatists weigh in too: will it understand relationships between files—especially code—or is this just fancy keyword vibes? Verdict from the crowd: cool demo, fierce debate. Some see a game‑changer for messy desktops; others see a cloud‑tethered privacy risk with drama on the side.
Key Points
- •Recall embeds images, audio, video, PDFs, and text into a local ChromaDB to enable semantic, cross‑modal search.
- •It uses Gemini Embedding 2 (768‑dim) for vectorization and cosine similarity for ranking results.
- •An animated setup wizard handles API key validation, folder selection, indexing, and Raycast configuration.
- •A Python API supports ingesting files/directories and performing natural-language searches with similarity scores.
- •A Raycast extension provides a grid UI with thumbnails and a headless command for instantly opening the best match.