dokosa

A command-line semantic search tool that indexes and searches local Git repositories using vector embeddings.

"dokosa" (どこさ？) is Japanese for "where is it?".

Features

• Semantic indexing: Uses OpenAI embeddings to create searchable vector representations of code
• Git integration: Automatically tracks repository commits and file changes
• Flexible filtering: Include/exclude files using glob patterns
• Chunked processing: Splits large files into overlapping chunks for better search granularity
• Similarity search: Find code snippets based on semantic meaning, not just keyword matching

Installation

$ cargo install dokosa

Quick Start

# Set your OpenAI API key
$ export OPENAI_API_KEY="your-api-key-here"

# Set the index file path
$ export DOKOSA_INDEX_FILE="$HOME/.dokosa"

# Add a repository to the index
$ dokosa add /path/to/your/repo

# Search for code semantically
$ echo "function to parse JSON" | dokosa search

# Sync repositories with latest commits
$ dokosa sync

Commands

• add - Index a Git repository for semantic search
• search - Find semantically similar code chunks
• list - Show all indexed repositories
• sync - Update repositories with latest changes
• remove - Remove a repository from the index

Run dokosa <command> --help for detailed options.