Matrix AI Assistant

This is a small bot for matrix which brings chatgpt functionality into any matrix room so that it can be collaboratively used.

Once the bot is invited to a room, it will join and wait until it is mentioned in a message. The bot supports multiple sessions just like the chatgpt webinterface and saves session history in a SQLite database. Sessions can be managed with different commands (see examples).

Please note:

• Currently it only works in unencrypted rooms.
• This is hacky, I created it in a few hours.

Examples

Here are some example screenshots. The help command is used to get a list of available commands.

With no commands, the bot will respond with an answer generated by the openai API.

The new command starts a session.

There are multiple commands to manage the sessions.

Install & run

Poetry is used for dependency management.

Install dependencies:

poetry install

Provide environment variables or a .env file with these variables:

# Set to true to run productively.
PRODUCTION=false
# Path to an sqlite db. If it doesn't exist, it will be automatically created!
DB_PATH="./mydb.db"
# Matrix homeserver url.
MATRIX_HOMESERVER="https://my.matrix.home.server"
# User name of the user used as the bot.
MATRIX_BOT_USERNAME="ai-assistant"
# An access token fot the user.
MATRIX_ACCESS_TOKEN="syt_MY_SECRET_TOKEN"
# An OpenAI API access key
OPENAI_API_KEY="my-secret-openai-api-key"

(More configuration options will be added in the future.)

Finally, run the bot:

poetry run matrix-ai-assistant

Roadmap

This is written in python. Currently the code base is messy and I want to rewrite everything in golang. See the golang branch.

Here is a list of possible future improvements:

• Restructure codebase, add tests and more documentation.
• Containerize the application.
• Replace/stop using simplematrixbotlib and just use nio directly.
• Support encrypted rooms.
• Handle things like connectivity loss to matrix server better.
• More configuration options.
• Introduce a usage quota per room to better control API fees.
• Save users in database and introduce a usage quota per user.
• Better control over the openai API.
  • Better control regarding token limit.
  • Better error handling (currently hard to do with the openai python package).
• Better control over the context and how much history is sent to the API.
• More modularized way to interact with API in order to support more/different APIs (Idea here is to let one swap out the API or model used).
• Different database schema where active session is not based on timestamp.