find anything, locally.

search names, paths, and file contents instantly. no cloud. no sync. just speed.

get started
coming soon

download in progress.

Builds for Windows and macOS are finishing final QA. Drop your email to be first in line.

windows
macos

v.0.41 · stable build · dropping soon

core infrastructure

instant lexical search

avg latency

12ms

old way

folder → folder → folder → file

minvector way

search → file

sub-50ms
lookup

Built on Tantivy, our engine maintains an atomic background inverted index. It scans and tokenizes file contents dynamically without locking threads or consuming idle system RAM.

minvector
search...
Projects
Downloads
Design Assets
·report.pdfpdf
·notes.mdmd
·main.tsts
·styles.csscss
·index.htmlhtml
supported types
documents
txtmdpdfdocxpptxhtml

* Converts raw PDF, DOCX, and HTML into structured markdown before indexing.

code
pytsrsgojsonyaml

* Captures function headers, block tags, and syntax structures seamlessly.

100%
local.

zero cloud dependencies. zero data leakage.

your files never leave your machine.

engine architecture

rust core runtime

Powered by Tokio async tasks and native OS file-system wrappers (notify + ignore). Natively honors .gitignore profiles and routes raw file extractions across a thread-safe parallel processing pool.

launch anywhere
+Space

one shortcut.

Hit the global hotkey from any app. minvector pops up in under 40ms. Search. Jump. Done.

memory footprint

~18MB

Idle RAM usage across a 200k file index. No background daemons. The watcher only activates on filesystem events.

live index
watching 4 directories

always fresh.

The file watcher uses OS-native events (inotify/FSEvents/ReadDirectoryChangesW) to update the index the instant a file is saved. Zero polling overhead.

developer api

plug in anywhere.

$ curl localhost:7700/search
-d '{"q":"design assets"}'

Local REST API + MCP server. Pipe results into any AI agent, script, or IDE plugin.

benchmark · full-text search on 100k files
minvector
12ms
spotlight (macos)
140ms
windows search
210ms
ripgrep (raw)
320ms
privacy model

air-gapped by design.

no outbound network calls
no telemetry or analytics
no login required
no account, ever

find anything.

read documentation