Skip to main content

​
Purpose

Use semantic search when you can describe the library but cannot guess the exact keyword or topic spelling.
List search on Semantic, with relevance percentages

Screenshot TODO · images/screenshots/search/semantic.webp

​
Requirements

  1. Pro is active
  2. A working Embedding provider / model in Settings → AI
  3. Build or refresh the AI index when the UI asks
  4. Switch list search to Semantic
Non-Pro sees a lock and the paywall. Without an index, results are empty or obviously thin.

​
Scope

Typical options:
  • starred: your Stars
  • knowledge: the knowledge base
  • all: broader (follow the app)
If you only want curated ingested content, prefer knowledge, or ask in the RAG workspace.

​
How it works (user-facing)

The query is encoded as a vector and compared with cached repo vectors. Some paths fuse with keyword ranking. Quality tracks the embedding model and whether the index covers the target repos. After you tap Build, a busy ring appears first; already-fresh repos advance the progress ring, and only missing or stale vectors are filled. Relevance percentages in the list match the type size of language, Stars, and Forks. Knowledge-base keyword search supports Chinese infix matching; naming a repo is more likely to hit indexed README content. See Index management.

​
Notes

  • Semantic search is not Search Center’s Web scope.
  • Changing the embedding model can misalign old vectors; rebuild when the UI says so.
  • Whether private repos enter the vector index depends on login scope and ingest range.
Last modified on August 25, 2026