Unlock the secrets of any codebase in seconds. The Codebase Onboarding Agent is a high-fidelity, autonomous intelligence engine designed to transform massive, complex GitHub repositories into structured, senior-developer-level onboarding reports.

• 🧠 Autonomous DAG Orchestration: Powered by LangGraph, the agent executes a sophisticated, multi-stage Directed Acyclic Graph (DAG) to decompose codebases logicallly.
• 🌳 Polyglot AST Extraction: Utilizes Tree-sitter for lazy-loading and deep structural analysis of Python, JavaScript, and TypeScript.
• 🔍 Local Semantic Memory: Implements ChromaDB with local ONNX-backed embeddings (all-MiniLM-L6-v2) for instant, private, and free semantic code search.
• ⚡ Real-time Pipeline Streaming: A modern FastAPI backend streams live analysis progress directly to the UI via WebSockets.
• 💅 High-Fidelity Dashboard: A gorgeous, dark-mode SPA built with Tailwind CSS, featuring dynamic metrics, interactive module maps, and architecture visualizations.
• 🤖 Infinite Context via OpenRouter: Seamlessly routes through 100+ models, currently optimized for Qwen-3 Coder to handle enterprise-scale repositories.

The agent operates through a strictly ordered execution pipeline, ensuring information scales from raw structural data to high-level architectural insights.

graph TD
    A[User Inputs GitHub URL] -->|WebSocket| B(FastAPI Server)
    B --> C{LangGraph Workflow}
    
    subgraph "Agent DAG Sequence"
    C --> D[parse_structure]
    D --> E[identify_tech_stack]
    E --> F[find_entry_points]
    F --> G[summarize_modules]
    G --> H[trace_data_flow]
    H --> I[extract_caveats]
    I --> J[compile_report]
    end
    
    J -->|JSON State| K["Tailwind Dashboard"]
    D -.->|Sync| L[("ChromaDB Vector Store")]
    L -.->|REST| M["/api/search"]

• Python 3.10+ (Required for LangGraph and Async FastAPI)
• Node.js / NPM (Required for Tailwind CSS compilation)
• Git (Installed on system PATH)

git clone https://github.com/KindaJayant/codebase-onboarding-agent.git
cd codebase-onboarding-agent
pip install -r requirements.txt

Create a .env file in the root directory:

# Get yours at https://openrouter.ai/keys
OPENROUTER_API_KEY=your_openrouter_api_key_here

# Build Tailwind CSS (v3)
npx tailwindcss -i ./static/input.css -o ./static/output.css

# Launch the Backend Engine
uvicorn main:app --reload

Visit http://localhost:8000 and begin your first analysis.

• main.py: The central FastAPI gateway handling HTTP, WebSockets, and environment orchestration.
• agent/:
  • graph.py: Defines the LangGraph topology and state logic.
  • nodes.py: The implementation layer for each LLM-backed analysis stage.
  • prompts.py: Highly-tuned system instructions for senior developer personas.
• utils/:
  • repo.py: Robust Git management and local OS temp buffering logic.
  • parser.py: Multi-language Tree-sitter AST extraction utilities.
  • vectorstore.py: Local ChromaDB indexing and ONNX embedding logic.
• static/: The Vanilla JavaScript SPA, including Tailwind source and compiled outputs.
• db/: Locally persistent vector index directory.

By default, this agent uses all-MiniLM-L6-v2 via the chromadb default embedding function. This runs completely on your CPU, ensuring that your codebase structure is never sent to a cloud embedding API, preserving your privacy and OpenRouter quotas.

The nodes.py engine features an Exponential Backoff retry loop. If OpenRouter's free-tier proxies hit a 429 Rate Limit, the agent will gracefully pause for 15 seconds and automatically resume, preventing UI crashes.

MIT License. See LICENSE for details. Developed with ❤️ for the global open-source community.