rhizome is a fast, minimal, single-binary command-line interface (CLI) tool designed for local-first semantic code search across multiple code repositories. It leverages natural language queries, Tree-sitter grammars for syntactically aware chunking, and an embedded vector database (SQLite with sqlite-vec) for efficient storage and retrieval. The user experience aims to be familiar to users of tools like ripgrep.

• Fast, Local-First Search: All indexing and searching occur on the user's machine.
• Single Binary: Easy distribution and installation.
• Semantic Search Focused: Prioritizes natural language queries.
• Minimal Configuration: Sensible defaults with a clear configuration file for customization.
• Ripgrep-like UX: Familiar command structure, output formatting, and exit codes.
• Robust Indexing: Reliable change detection and transactional database updates.

• Indexes code repositories locally.
• Performs semantic search using natural language queries.
• Supports structure-aware chunking for Rust, Python, and TypeScript using Tree-sitter.
• Provides fallback text-based chunking for other UTF-8 files.
• Uses ONNX embedding models (defaults to BAAI/bge-small-en-v1.5).
• Stores index in an SQLite database (~/.rhizome/index.db) with vector support via sqlite-vec.
• Respects .gitignore rules during indexing.
• Provides commands for initialization, indexing, searching, status checking, and deleting index data.
• Offers configurable options via ~/.rhizome/config.toml.
• Ripgrep-like output formatting with optional scores and context display.

• rhizome init [--force]: Initializes configuration (~/.rhizome/config.toml), creates necessary directories (~/.rhizome/models/), initializes the database (~/.rhizome/index.db), and downloads the default embedding model. Use --force to overwrite existing configuration.
• rhizome index [--repo <path>] [--dry-run] [--include-ignored] [--ext <extensions>] [--jobs <N>]: Indexes the specified repository (or current directory). Discovers files, chunks content, generates embeddings, and stores data in the database.
  • --dry-run: Shows what would be processed without modifying the database.
  • --include-ignored: Includes files normally ignored by .gitignore.
  • --ext: Comma-separated list of file extensions to include.
  • --jobs: Number of parallel jobs for processing (defaults to number of CPUs).
• rhizome search <query> [--repo <path>] [--limit <N>] [--show-scores] [--show-range] [--context <N>] [--path <glob>] [--lang <lang>]: Performs a semantic search using the provided query.
  • --limit: Maximum number of results.
  • --show-scores: Displays similarity scores.
  • --show-range: Shows the start and end line numbers of the chunk.
  • --context: Number of context lines to display before and after matches.
  • --path: Filter results to files with paths matching these patterns.
  • --lang: Filter results to a specific programming language.
• rhizome status [--repo <path>]: Shows the indexing status for a repository, including indexed file/chunk counts, languages, and detected changes (new, outdated, deleted files).
• rhizome delete [--repo <path>] [--all] [--yes]: Removes index data. Can remove data for a specific repository or the entire index (--all). Requires confirmation unless --yes is provided.

Global Flags:

• --verbose: Enable verbose output.
• --help: Show help message.
• --version: Show version information.
• --no-color: Disable colored output.

Configuration is managed via ~/.rhizome/config.toml. Key options include:

• [model]:
  • path: Identifier for the Hugging Face model (e.g., "Xenova/bge-small-en-v1.5") or a local path.
  • provider: ONNX runtime provider (e.g., "cpu").
  • dimension: Embedding dimension (e.g., 384).
• [chunking]:
  • size: Target chunk size in tokens (e.g., 450).
  • overlap: Ratio of overlap between chunks (e.g., 0.1).
• [database]:
  • path: Path to the SQLite database file (e.g., "~/.rhizome/index.db").
• [paths]:
  • models_dir: Directory to store downloaded models (e.g., "~/.rhizome/models").

The init command creates a default configuration file if one doesn't exist.

Install ONNX Runtime and system dependencies locally by running:

./scripts/setup_deps.sh

This script downloads the runtime and copies its libraries into target/debug. Run tests with:

ORT_STRATEGY=system ORT_LIB_LOCATION=target/debug make test

The project uses a Makefile for common development tasks:

• make check: Verify formatting and linting.
• make test: Execute tests.
• make run-with-args ARGS='...': Run the application with specific arguments.
• make all: Run checks and tests together.