/workgpt

QA LLM from Confluence data

Primary LanguagePythonMIT LicenseMIT

WorkGPT

QA over Documents pipeline with a focus on privacy. All models and logic run locally. Suitable for use with company data.
Implemented on LangChain and with a simple Streamlit WebUI.
Reads from a private Confluence site, calculates embeddings with HuggingFaceEmbeddings and generates responses with the GPT4All LLM.
Dockerized for easy deployment.

$ docker run [...] workgpt 
Error: Please provide a command.
Usage: ./entrypoint.sh <command>
Available commands:
  webui        Runs streamlit webserver
  rebuild-db   Creates a new vector store and loads Confluence pages
  append-db    Appends Confluence pages to existing vector store
  interactive  Runs in interactive mode

Configured via environment variables; see .env.example.

example

Pending / TODO:

  • Broken concurrency; move to langhcain ts
  • Tweak text splitting and chunk sizes
  • Test MultiQueryRetriever