Summary

CodeSeek is a Rust-powered CLI tool providing advanced code intelligence for AI coding agents. It builds call graphs and hybrid semantic search indexes across seven languages, enabling sophisticated terminal-based code analysis and integration with tools like Claude Code and Codex CLI.

How It Works

Indexing (codeseek init) uses Tree-sitter for AST parsing, extracts code entities, generates embeddings via API (LanceDB), builds BM25 indexes (Tantivy), and serializes call graphs (PetCodeGraph). Incremental updates use file MD5 hashes, with optional git hooks for auto-re-indexing. The hybrid search pipeline combines dense vector search, sparse keyword search, and graph traversal. Results are fused via Reciprocal Rank Fusion (RRF) and re-ranked by a cross-encoder model, with a fallback to graph-based name search.

Quick Start & Requirements

Installation is via npm (npm install -g codeseek), Homebrew (brew install codeseek), or source build. Requires Node.js/npm and Rust (for source). An interactive wizard configures embedding models (API token, base URL). Supported platforms: macOS (arm64, x64), Linux (x64).

Highlighted Details

• Supports 7 languages: Rust, Python, JS/TS, Go, C/C++, Java.
• Hybrid search: Dense (LanceDB), Sparse (Tantivy BM25), Graph (PetCodeGraph).
• Advanced ranking: RRF fusion and Qwen3-Reranker-4B cross-encoder.
• Direct integration with Claude Code/Codex CLI for terminal call graph tracing and symbol search.
• Incremental indexing via MD5 hashes and optional git hooks.

Maintenance & Community

The README provides no specific details on maintainers, community channels, sponsorships, or roadmap.

Licensing & Compatibility

Released under the MIT license, permitting commercial use and integration into closed-source projects.

Limitations & Caveats

Currently supports only macOS and Linux (x64), excluding Windows. Full semantic search functionality depends on external embedding and reranking models, potentially requiring API keys and incurring costs.