#Semantic Search in Basic Memory

You wrote a note three weeks ago about “hardening the authentication system.” Today you search for “login security improvements.” No results.

The words are different. The meaning is the same. That’s the problem with traditional search. It aims to match words, not concepts. If you don’t remember exactly how you phrased something, you might never find it again.

As of v0.19.0, Basic Memory works differently. It understands meaning. Search for “login security improvements” and it finds your note about authentication hardening, because it knows those two things are about the same idea.

It’s the difference between a filing cabinet and a librarian. A filing cabinet only gives you what you put in the right folder. A librarian listens to what you’re looking for and finds it.

How to Think About It

When you save a note, Basic Memory does two things: it indexes the words (like traditional search always has), and it also encodes the meaning of the text.

When you search, both of those indexes get checked. If something matches on words and meaning, it rises to the top. If it only matches on meaning (like your auth hardening note showing up for a login security query) it still comes back.

What This Looks Like in Practice

• “ways to make the frontend faster” finds notes about React performance, bundle size, lazy loading — even if none of them use the word “faster”
• “what we decided about the database” finds notes about your Postgres vs SQLite evaluation, schema decisions, migration planning
• “that conversation about pricing” finds notes about pricing strategy, customer feedback on tiers, competitive analysis

You just search the way you think.

Semantic search docs →

v0.19.0 release →