Skip to main content

Search

Lektr provides powerful semantic search across your entire highlight library.

How It Works​

Lektr uses hybrid search combining:

  • Vector similarity (semantic/AI-based)
  • Full-text search (keyword matching)
  • Reciprocal Rank Fusion (RRF) to combine results

This means you can search for concepts, not just exact words.

  1. Navigate to Search
  2. Enter your query
  3. Results show matching highlights with:
    • Highlighted matching text
    • Book title and author
    • Relevance score

Search Tips​

Conceptual Searches​

Search for ideas, not just keywords:

  • "dealing with failure" finds highlights about resilience, setbacks, learning from mistakes
  • "time management" finds productivity-related highlights

Multi-word Queries​

Use natural language:

  • "how to build habits"
  • "importance of reading"

Author/Book Searches​

Include book or author names:

  • "Atomic Habits routines"
  • "Seneca stoicism"

Technical Details​

Vector Embeddings​

Highlights are embedded using AI models and stored in PostgreSQL with the pgvector extension. This enables semantic similarity search.

Search Pipeline​

Query → Embedding → Vector Search ─┐
                                    ├─→ RRF → Ranked Results
Query → Text Search ───────────────┘

Re-generating Embeddings​

If you modify embedding settings, regenerate via:

curl -X POST /api/v1/admin/embeddings/regenerate \
  -H "Cookie: auth_token=TOKEN"