This project provides an ID-based RAG (Retrieval-Augmented Generation) API using FastAPI, Langchain, and PostgreSQL/pgvector. It's designed for scalable document indexing and retrieval, particularly for use cases requiring file-level embedding management, such as integration with LibreChat. The API enables targeted queries by leveraging file metadata stored in a database.

How It Works

The core approach is to organize document embeddings by file_id, enabling granular control and targeted retrieval. It utilizes Langchain's vector store capabilities for efficient searching and supports asynchronous operations for enhanced performance. The architecture is built with FastAPI, offering a modern and scalable web framework.

Quick Start & Requirements

• Install: pip install -r requirements.txt
• Prerequisites: PostgreSQL with pgvector extension (v0.5.0+ recommended), Python 3.x. OpenAI API key is required for default embeddings.
• Run: Use docker compose up for a combined RAG API and PostgreSQL/pgvector setup, or run the API separately with docker compose -f ./api-compose.yaml up. Local setup requires configuring a .env file and running uvicorn main:app.
• Docs: README

Highlighted Details

• Supports multiple embedding providers including OpenAI, Azure OpenAI, Hugging Face (via sentence-transformers or TEI), and Ollama.
• Offers optional authentication via JWT verification.
• Allows configuration of chunking strategy, including chunk size and overlap.
• Can utilize Atlas MongoDB as an alternative vector database.

Maintenance & Community

The project is maintained by danny-avila. Further community or roadmap information is not detailed in the README.

Licensing & Compatibility

The README does not explicitly state a license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is described as a "simple API" and may evolve, suggesting it might be in early development. Specific limitations regarding supported document types beyond PDFs or advanced RAG techniques are not detailed.