Skip to content

CLAUDE.md

Latest commit

 

History

History
37 lines (30 loc) · 1.04 KB

CLAUDE.md

File metadata and controls

37 lines (30 loc) · 1.04 KB

Knowledge Vector Search - Claude Code Instructions

Project Overview

ONNX-optimized vector search system for knowledge bases. Uses FastEmbed (23MB) + sqlite-vec for semantic search across documents with incremental updates.

Tech Stack

  • Python 3.12+
  • FastEmbed - ONNX embeddings (CPU-optimized)
  • sqlite-vec - Vector similarity in SQLite
  • PyYAML - Frontmatter parsing

Development Commands

# Environment
uv sync

# Testing & Quality
python -m pytest
ruff format .
mypy knowledge_search/

# Usage
python -m knowledge_search.embed /path/to/docs
python -m knowledge_search.search "query" --limit 5
python -m knowledge_search.smart_search "query" --limit 10

Key Components

  • KnowledgeSearch class - main API
  • Incremental processing via SHA256 change detection
  • Metadata extraction from YAML frontmatter
  • ~300ms query performance, ~6KB per document storage

Architecture Notes

  • No external DB required (SQLite-based)
  • Optimized for Obsidian/Markdown knowledge bases
  • Smart search wrapper with auto-index updates