---
title: "How search works"
url: https://memocore.ai/memos/6a55d6f4fcb359b0f635da39
project: "Memocore Knowledge Base"
tags: "features"
updated_at: "2026-09-14T04:19:29.375Z"
---

# How search works

Search is how your AI finds the right memory before it answers - so the relevant note surfaces instead of a keyword dump. You do not usually run search yourself; your connected AI searches your memory automatically, but you can also search in the dashboard.

What it gives you:
- It finds a memo by meaning as well as by wording. Every memory is turned into an embedding - a vector capturing what it is about - and so is your query, so the two can be compared by meaning; that is combined with ordinary keyword matching over the same query, and a memo found by both sides ranks highest. This is why a note comes back even when you ask for it in words you did not use when you saved it.
- Results are ranked by relevance, weighting a memo's internal retrieval cues most heavily, then its title, preview and content. That is why filling a memo with good cues makes it easy to find later.
- It tolerates small typos, so a slightly misspelled query still finds the note.
- Results are lightweight: title, a short preview, whether the memo has files, and where it lives (project, company and tags by name). The full content is fetched only for the memos that matter.
- Every search comes with a plain-language verdict on how solid the results actually are: which memos are genuinely on topic, which merely share a word, and when the honest answer is that your memory holds nothing on the subject. A search always returns a page of nearest memos, so that verdict is what stops an assistant from trusting a lookalike.

Scope: search can span everything you can access (your private memos plus all your projects), or be narrowed to just your private memos, a single project, or one company's memory. You can also filter by tags - keep only memos carrying all the tags you list, or drop the ones carrying tags you exclude.

Duplicate detection: when you save something, Memocore checks for near-duplicate memos and nudges you to merge rather than create a second copy.

Fair use: preparing your memos for search, and running searches, both come with usage allowances that depend on your plan. Normal use stays well within them - they are only there to keep things fair. A search run within a company, or within one of its projects, draws on that company's allowance; otherwise it counts toward your own. The exception is a project you only follow - a public project you joined, or a private one you were invited to: your searches there count toward your own allowance, not the project owner's.