treeseek

CLI com daemon watcher para busca semântica de código. Indexa codebases com AST-aware chunking, gera embeddings via Ollama e armazena em LanceDB local — 100% offline, gratuito e privado.

Prerequisites

1. Node.js 20 or 22 (do NOT use Node 24 — incompatible with tree-sitter)
2. Ollama running locally: https://ollama.com

   ollama serve

3. Embedding model pulled:

   ollama pull nomic-embed-text

Installation

npm install -g treeseek

Commands

treeseek index <path>              # Index codebase (full or incremental)
treeseek search <query> <path>     # Semantic search
treeseek status <path>             # Check indexing status
treeseek clear <path>              # Remove index
treeseek watch <path>              # Start watching and auto-reindexing
treeseek stop [path]               # Stop watcher (--all for all projects)
treeseek list                       # List watched projects

All commands support --json for machine-readable output.

Examples

# Index and start watching (daemon in background)
treeseek watch /my-project

# Search code semantically
treeseek search "authentication JWT" /my-project
treeseek search "database connection" /my-project --top-k 5 --json

# Check status
treeseek status /my-project

# Stop watching
treeseek stop /my-project

How It Works

1. AST-Aware Chunking — Tree-sitter parses code in 9 languages (TS, JS, Python, Java, Go, Rust, C++, C#, Scala), splitting by functions, classes and methods
2. Ollama Embeddings — Generates vectors locally with nomic-embed-text (768 dim). No external APIs
3. LanceDB — Stores vectors locally in ~/.codebase-indexer/. Cosine similarity search
4. Merkle DAG — Incremental indexing detects changes via SHA-256 hashes. Only reindexes modified files
5. Daemon Watcher — treeseek watch monitors file changes with 10s debounce and auto-reindexes

Environment Variables

| Variable | Default | Description | |---|---|---| | OLLAMA_HOST | http://localhost:11434 | Ollama API endpoint | | EMBED_MODEL | nomic-embed-text | Ollama embedding model name | | LANCEDB_PATH | ~/.codebase-indexer | Base directory for LanceDB storage | | EMBEDDING_BATCH_SIZE | 100 | Batch size for embedding generation | | CUSTOM_EXTENSIONS | — | Additional file extensions (comma-separated) | | CUSTOM_IGNORE_PATTERNS | — | Additional ignore patterns (comma-separated) |

Supported Languages

TypeScript, JavaScript, Python, Java, Go, Rust, C, C++, C#, Ruby, PHP, Swift, Kotlin, Scala, Markdown, Jupyter Notebooks.

Architecture

Codebase → Tree-sitter AST → AST Chunker → Ollama Embeddings → LanceDB
                                                           ↓
                                              treeseek search (CLI)
                                                           ↓
                                              Cosine Similarity → Results

Daemon Watcher

treeseek watch <path> starts a background process that:

• Performs initial indexing if not already indexed
• Monitors file changes with fs.watchFile (10s debounce)
• Auto-reindexes only modified/added files
• Manages PIDs in ~/.treeseek/pids/
• Logs to ~/.treeseek/logs/

treeseek watch /my-project    # Start daemon
treeseek list                  # List watched projects
treeseek stop /my-project      # Stop specific project
treeseek stop --all            # Stop all watchers