penn-cis-chatbot

Basic LLM chatbot for answering internal Q&A. This is a simple langchain chatbot that (1) integrates with Slack, (2) indexes a set of documents using ChromaDB, (3) interfaces with GPT 3.5 to answer questions. it remembers (limited) chat history.

Setup

• Run pip3 install -r requirements.txt to configure libraries
• Create a subdirectory called index_these and copy Word + PDF documents into it
• Create a Slack application in your Workspace. Ensure it has permissions for at least:
  • im:write (direct messages)
  • mpim:write (group messages)
  • app_mentions:read (read @ messages)
  • chat:write (send messages as itself)
  • channels:join (join public channels)
  • channels:history (view messages in public channels)
  • channels:read (view basic info about public channels)
  • files:read (read shared files)
  • im:history (read history in direct messages)
  • dnd:read (read do-not-disturb settings)
  • users:read (know users in workspace)
  • Also, be sure to enable Socket Mode and Enable Events -- otherwise, the app won't know about messages
• Set up environment variables:
  • SLACK_BOT_TOKEN (from creating the Slack app; should start with xoxb)
  • SLACK_APP_TOKEN (from creating the Slack app; should start with xapp)
  • OPENAI_API_KEY (from your OpenAPI subscription for GPT)

Running

Run python3 slackbot.py to launch the chatbot. It will first index all of the contents of index_these, then join Slack.