Codie is a powerful CLI tool designed to help developers quickly understand and navigate codebases. Whether you're onboarding a new project or exploring an unfamiliar repository, Codie provides AI-powered insights to accelerate your workflow.

• Go 1.18 or higher
• OpenAI API key (you'll be prompted to provide this on first run, or you can set it in a .env file)

# Clone the repository 
git clone https://github.com/exolottl/codie
cd codie

# Install dependencies
go mod tidy

Create a .env file in the project root with your OpenAI API key:

OPENAI_API_KEY=your_api_key_here

Before generating a summary, you need to index your codebase:

go run main.go index <directory path>

This scans your codebase, processes code files, and generates embeddings using OpenAI's API. The embeddings are saved to embeddings.json for future use.

After indexing, you can generate a summary of the codebase:

go run main.go summarize <directory path> [options]

Options:

• --detail=<level> - Set detail level (brief, standard, comprehensive)
• --focus=<path> - Focus on a specific directory
• --no-metrics - Exclude code quality metrics

For backward compatibility, running just go run main.go <directory path> will perform the indexing operation.

1. Code Scanning: Codie scans your codebase for supported file types (.py, .js, .go, etc.)
2. Smart Chunking: Files are broken into meaningful semantic chunks using Tree-sitter parsers for better analysis
3. Efficient Batch Processing: Code chunks are processed in batches through OpenAI's embedding API for optimal performance
4. AI Embeddings: Generated embeddings capture the semantic meaning of your code
5. AI Analysis: Codie builds a prompt based on your code and uses OpenAI's models to generate insightful summaries
6. Structured Output: Summaries include overview, architecture, key features, and implementation details

• Code Repo Indexing – Process and embed code for AI analysis
• Code Repo Summarization – AI-powered summaries for understanding large codebases
• Focused Analysis – Zoom in on specific directories or parts of your codebase
• Configurable Detail Levels – Choose between brief, standard, or comprehensive summaries
• Syntax-Aware Parsing – Uses Tree-sitter to understand code structure for smarter analysis

• Functionality Finder – Locate specific functions or features within a project
• Code Requests – Ask Codie for code snippets, explanations, and recommendations

Codie uses a combination of technologies to provide intelligent code analysis:

• Tree-sitter for language-specific parsing of code structures (functions, classes, methods)
• Anthropic SmallEmbedding3 model for semantic code processing
• OpenAI's GPT models for generating the final analysis
• Concurrent processing for efficient handling of large codebases
• Rate limiting to manage API usage and prevent throttling

• github.com/charmbracelet/glamour - For Markdown rendering in terminal
• github.com/schollz/progressbar/v3 - For progress visualization
• github.com/sashabaranov/go-openai - OpenAI API client
• github.com/smacker/go-tree-sitter - Code parsing and analysis
• Tree-sitter language parsers for Go, JavaScript, Python, and more

MIT License