CocoSearch

关于

CocoSearch is a privacy-first, local-first hybrid semantic code search tool designed to help developers quickly understand and navigate codebases. It combines vector similarity from local Ollama-generated embeddings with traditional keyword matching using RRF fusion, allowing users to find code by its meaning, not just exact text. Leveraging CocoIndex and Tree-sitter for intelligent code chunking and symbol extraction, and PostgreSQL with pgvector for efficient storage, all operations remain entirely on your machine. It offers versatile interfaces including a command-line interface (CLI), an interactive REPL, a web dashboard, and seamless integration as an MCP server for AI assistants like Claude Code.

主要功能

• Hybrid search combining semantic similarity and keyword matching via RRF fusion.
• Symbol filtering to narrow results by functions, classes, methods, or interfaces with glob patterns.
• Context expansion that automatically expands results to enclosing function or class boundaries.
• Query caching for faster repeated searches with exact and semantic thresholds.
• Privacy-first design with all operations running locally, no external API calls or telemetry.
• 0 GitHub stars

使用案例

• Accelerate self-onboarding to new codebases by quickly finding relevant code sections.
• Integrate with AI assistants (e.g., Claude Code) for enhanced code exploration and generation.
• Perform advanced code analysis, debugging, and pipeline diagnostics to understand search result rankings.