Intelligence

Semantic Search

Find by meaning, not keywords. Search across every document, even when the words don't match.

Why it matters

Ctrl-F finds exact words. But "a clause that limits our liability" can be written ten different ways. Semantic search understands meaning, not spelling, and it runs on local embeddings, so your query never leaves your device, even as it reads across your whole library.

How it works

From a question to the right passage

1

Toggle search

A sparkle button flips search into semantic mode whenever you want meaning over exact match.

2

Ask naturally

Type a concept or a question, not the precise phrase you're hoping to match.

3

Local embeddings

A multilingual model (bge-m3) scores similarity entirely on your machine.

4

Ranked results

The closest passages surface first, from every document, not just the open one.

What you can do

Search that understands

Natural-language queries

Search a concept, not a keyword. Phrase it the way you'd ask a colleague.

Across every document

One query spans your whole library, surfacing matches wherever they live.

Local embeddings

The query and the index stay on-device, so nothing is sent anywhere to search.

Multilingual

bge-m3 handles Polish and other languages, built for real legal and formal text.

See it in action

It works in your language

Search by meaning, privately

Local embeddings, on your machine, so your search queries never leave your device. Free, always.