semantic - local - on-device embeddings
private semantic search for macOS
Search your Mac the way you remember things.
Memsearch turns files, photos, notes, and PDFs into a local memory index. Describe the thing you remember and get back the file that matches.
local indexbring your own keyopen source
01How it works
Index. Embed. Search. Nothing else needs an account.
A background indexer reads files on disk, asks Gemini to summarize and embed them, then writes vectors to local storage. The app is designed to feel like a fast command palette, not a cloud drive.
STEP 01
Point it at folders
Choose Documents, Desktop, screenshots, a repo, or a Photos library. Memsearch reads files on disk and keeps the index local.
$ watch ~/Documents
$ watch ~/Pictures
$ watch ~/Screenshots
STEP 02
Build the local index
Files are summarized and embedded with your Gemini key. Face tags and vectors are written to ChromaDB on your Mac.
-> extract text + thumbnails
-> embed 1536d vectors
-> chromadb.add(batch)
STEP 03
Search by memory
Type what you remember: "whiteboard from standup", "neon alley tokyo", or "@mira beach sunset".
query -> embed -> topK
@mira + "beach sunset"
0.94 IMG_4821.HEIC
02vs. Spotlight
Finder searches filenames. Memsearch searches meaning.
Dimension
Spotlight / Finder
Memsearch
Match by
Exact filename, metadata
Meaning of file contents
Photos
Surfaced only if named
Described naturally
People
By filename tag
@person via on-device face detection
PDF bodies
OCR, keyword only
Semantic summary plus embedding
Result ranking
Filesystem heuristics
Vector similarity plus recency
Data leaves device
Cloud sync by default in many workflows
Index stays local
03Privacy
Your index. Your disk. Your call.
01
Local index
ChromaDB lives on your Mac. Nothing syncs in the background.
02
Your API key
Embedding calls use the key you provide during setup.
03
Open source
Inspect the app, build it yourself, or fork the workflow.
04
Off switch
Pause indexing and clear the index when you need a clean slate.
04Install
$0 - bring your own key
Free to download. You pay Google, not us.
Memsearch is free and open source. Paste your Gemini API key on first run. Embedding a typical personal library is a small one-time cost, and searches are near-free.