FLOAT

FLOAT is a powerful codebase chat assistant that runs entirely locally. It consists of a FastAPI backend and a Next.js frontend.

---

Features

Codebase Indexing & Retrieval

• AST-Based Semantic Chunking: Code-aware parsing for multiple languages.
• Hybrid Search Pipeline: Uses jinaai/jina-embeddings-v2-base-code vectors in ChromaDB, reranked via a BAAI/bge-reranker-base cross-encoder for precise context retrieval.
• Incremental Indexing: Syncs and refreshes repository changes automatically via the GitHub API.

Context-Aware Code Chat

• Codebase & File-Level Chat: Chat with your entire workspace using relevant context retrieval, or focus on a single file.
• Dependency Graph Injection: Maps cross-file dependencies, imports, and call chains to trace entry points and orchestrators.
• Flexible Model Support: Supports OpenAI, Gemini, Anthropic, DeepSeek, xAI, Groq, OpenRouter, and local models via Ollama.

Automated Documentation & Architecture Visualization

• Multi-Section Doc Generator: Auto-generates structured codebase documentation (Overview, Architecture, API Reference, Configuration).
• Mermaid Graph Visualizer: Generates interactive architecture diagrams (zoomable and pannable) directly from repository structure.

---

Setup & Quick Start

To configure your environment and run FLOAT, follow the Setup & Quick Start Guide.

---

Security & Storage Notice

Before running the application, please note:

• GitHub Personal Access Token (PAT): Stored locally in plaintext inside backend/data/.tokens.
• LLM API Keys (OpenAI, Gemini, etc.): Stored unencrypted in the browser's localStorage.
• Local-Only Operation: This tool is designed to run self-hosted. No credentials or repository data are uploaded to third-party servers.