Pro5 min read· Updated 2026-05-09

Semantic Search

Search by what you remember, not by exact keywords.

Semantic search is the difference between guessing the right keywords and describing what you remember. LinkVolv understands the meaning behind your library, not just the words on the page.

Why semantic search?

Keyword search fails on the queries that matter most: when you cannot remember the title, the author, or the exact terminology the page used. You usually remember what it was about — and that is exactly what semantic search matches against.

Compare these two queries

Keyword: react animation library

Semantic: “That React library someone tweeted last month for spring physics on lists”

The second query has zero exact-match keywords with the saved page. Semantic search still finds it.

How to search

Opening the search bar

Press ⌘ K from anywhere in the dashboard. On Windows / Linux: Ctrl K. The search palette opens on top of whatever you were doing.

Writing a good query

  • Use natural language. Full sentences work better than keywords. “The article about deep work and async teams” outperforms deep work async.
  • Describe context, not just topic. “The benchmark I saved while researching pgvector at 10M rows” works because the saved page mentioned both.
  • You can be approximate. Spelling, exact phrases, even the title — none of it has to match. The query is matched against the meaning of the page, not the surface form.

Filters and operators

Filters narrow a semantic query. They live next to the search input and accept the obvious shapes:

  • tag:vectors — restrict to bookmarks with that tag
  • session:"Postgres research" — restrict to a Session
  • since:7d / since:2026-04-01 — restrict by save date
  • has:archive — only bookmarks with an archived copy
example queries
text
# Last week's research on a topic
that benchmark with pgvector at 10M rows  since:7d

# A specific session, narrowed by intent
the postgres index trade-off  session:"Postgres research"

# Articles you can still read offline
incident postmortem with cascading failure  has:archive

How it works under the hood

At save time, every Bookmark’s content is embedded into a high-dimensional vector and stored in a vector database. When you search, your query is embedded with the same model and compared against the library by cosine similarity. The top results are re-ranked using the page summaries and any active filters before being shown to you.

Hybrid retrieval

For very short queries, LinkVolv blends vector similarity with keyword (BM25) scoring. This gives accurate results even for single-word searches, where pure semantic ranking can drift.

Free vs. Pro

  • Free — keyword search across titles, summaries, and key points.
  • Pro and Lifetime Plus — full semantic search with vector retrieval, hybrid ranking, and all filter operators.
Semantic Search — Documentation | LinkVolv