FINKI Hub / Chat Bot

RAG chat bot for the FINKI Hub Discord server, powered by LangChain and FastAPI. Uses PostgreSQL and pgvector for keeping documents. Has support for many LLMs.

It currently works on a dataset of documents (FAQ). It is planned to support other types of data as well.

Services

This project comes as a monorepo of microservices:

• API (/api) for managing documents, links and chatting (default port: 8880)
• GPU API (/gpu-api) for locally executing GPU accelerated tasks like embeddings generation (default port: 8888)
• Front-end (WIP)
• Database (PostgreSQL + pgvector) for keeping documents and embeddings

The API Docker image is available as ghcr.io/finki-hub/chat-bot-api, while the GPU API Docker image is available as ghcr.io/finki-hub/chat-bot-gpu-api.

Quick Setup (Production)

It's highly recommended to do this in Docker.

To run the chat bot:

1. Download compose.prod.yaml
2. Download .env.sample, rename it to .env and change it to your liking
3. Run docker compose -f compose.prod.yaml up -d

The API will be running on port 8880. This also brings up a pgAdmin instance. You may use it to view or create documents. It's accesible on port 5555 by default.

Quick Setup (Development)

Requires Python >= 3.13 and uv.

1. Clone the repository: git clone https://github.com/finki-hub/chat-bot.git
2. Install dependencies: in each directory (api and gpu-api), run uv sync
3. Prepare env. variables by copying env.sample to .env - minimum setup requires the database configuration, it can be left as is
4. Run it: docker compose up -d

This also brings up an OpenAPI instance at localhost:8880/docs.

Endpoints

This is an incomplete list. You may view all available endpoints on the OpenAPI documentation.

• /questions/list - get all questions
• /questions/name/<name> - get a question by its name
• /questions/embed - generate embeddings for all questions for a given model

License

This project is licensed under the terms of the MIT license.