This project is a chatbot for Mattermost that integrates with the Anthropic API to provide helpful responses to user messages. The chatbot - like this readme - is mostly written by Claude 3 AI, listens for messages mentioning the chatbot or direct messages, processes the messages, and sends the responses back to the Mattermost channel.

• Responds to messages mentioning "@chatbot" (or rather the chatbot's username) or direct messages
• Extracts images, PDFs and other files from message attachments and from URL links in messages
• Supports FlareSolverr to bypass Javascript/CAPTCHA restrictions
• Supports the Vision API for describing images. Images from PDFs will also be sent here.
• Gets transcripts of YouTube videos for easy tl;dw summarizations. Title, description and uploader are also provided
• Accesses additional live information via function calling. Currently supported: stock data (via Yahoo Finance, eg. ask about AAPL), cryptocurrency data ( via Coingecko), fiat currency exchange rates (via ECB)
• Maintains context of the conversation within a thread
• Sends typing indicators to show that the chatbot is processing the message
• Utilizes a thread pool to handle multiple requests concurrently (due to mattermostdriver-asyncio being outdated)
• Offers Docker support for easy deployment

• Python 3.11 or just a server with Docker (you can get away with using 3.8 if you use datetime.datetime.utcnow() instead of datetime.datetime.now(datetime.UTC))
• Anthropic API key
• Mattermost Bot token (alternatively personal access token or login/password for a dedicated Mattermost user account for the chatbot)
• The bot account needs to be added to the team and to the channels you want it to watch

1. Prepare a Mattermost bot account token and Anthropic API key

   • Create an Anthropic API key here
   • Create a Mattermost bot account token directly in the chat: Left menu -> Integrations -> Bot accounts. Give it the post:channels permission (this also suffices for DMs)

2. Clone the repository:

   git clone https://github.com/Elehiggle/Claude3MattermostChatbot.git
   cd Claude3MattermostChatbot

3. Install the required dependencies:

   pip3 install -r requirements.txt

   or alternatively:

   python3 -m pip install anthropic mattermostdriver certifi beautifulsoup4 pillow httpx youtube-transcript-api yt-dlp PyMuPDF defusedxml yfinance pymupdf4llm

4. Set the following environment variables with your own values:

Run the script:

python3.12 chatbot.py

The chatbot will connect to the Mattermost server and start listening for messages. When a user mentions the chatbot in a message or sends a direct message to the chatbot, the chatbot will process the message, extract content from links (if any), process images, PDFs and other files, handle image content using the Vision API, and send the response back to the Mattermost channel.

Note: If you don't trust your users at all, it's recommended to disable the URL/image grabbing feature, even though the chatbot filters out local addresses and IPs.

You can also run the chatbot using Docker. Use the following command to run the chatbot container:

docker run -d --name chatbotclaude \
  -e AI_API_KEY="your_ai_api_key" \
  -e AI_MODEL="claude-3-opus-20240229" \
  -e MATTERMOST_URL="mattermostinstance.example.com" \
  -e MATTERMOST_TOKEN="your_mattermost_token" \
  -e MAX_TOKENS="4096" \
  -e TEMPERATURE="0.15" \
  ghcr.io/elehiggle/claude3mattermostchatbot:latest

The official Mattermost AI plugin is a great way to integrate AI into your Mattermost server with its own advantages, like a nice UI. However, the features differ greatly. It requires direct access to the server, whereas this bot merely needs a bot token and can be conveniently used in a Docker environment.

Also there is:

• no fine granular control over certain settings (especially AI settings)
• no CAPTCHA bypass
• no Vision API
• no PDF parsing
• no YouTube transcript fetching
• and more

While the official plugin certainly will improve over time, this bot here will too and there will be certain features that will absolutely never make it into the official plugin, due to it primarily focusing on features for developers like function calling to retrieve GitHub issues, for example.

• Typing indicator is only sent to the channel, not the conversation thread. There is some issue I haven't figured out yet. I even prefer it this way, but mobile users can't see the channel while in a thread.

Other than that, while the chatbot works great for me, there might still be some bugs lurking inside. I have done my best to address them, but if you encounter any issues, please let me know!

Please note that the monkey patch in the code is necessary due to some SSL errors that occur because of a mistake within the mattermostdriver library. The patch ensures that the chatbot can establish a secure connection with the Mattermost server.

OpenAI ChatGPT Mattermost Chatbot

Contributions are welcome! If you find any issues or have suggestions for improvements, please open an issue or submit a pull request.

This project is licensed under the MIT License.

• Anthropic for providing the API for generating responses
• Mattermost for the messaging platform
• mattermostdriver for the Mattermost API client library
• chatgpt-mattermost-bot for inspiring me to write this python code
• youtube-transcript-api for the YouTube Transcript Fetch library
• yt-dlp for the YouTube API that allows us to fetch details
• PyMuPDF for the comprehensive PDF parsing library
• beautifulsoup4 for the HTML parsing library
• pillow for the image processing library
• httpx for the HTTP client library
• certifi for the certificate verification library