AbbyLarby is a Slack bot for connecting to generative AI helpers.

Available commands:

• /ai <prompt>: Show the OpenAI text API's response to <prompt>. Visible only to you.
• /ai img <prompt>: Show the Dall-E image generated by <prompt>. Visible only to you.
• /ai say <prompt> or * /ai say img <prompt>: immediately post the result to the channel.

Have a longer conversation by talking directly with the bot in the app Messages tab.

The bot can also be added to any channel, and then will respond to @ mentions.

Configuration is via the environment variables set in .env.example. You can copy the file to .env or set environment variables another way.

• MODEL: a model using the OpenAI chat API, e.g. "gpt-4" or "gpt-3.5-turbo".
• MODEL_MAX_TOKENS: the maximum number of tokens allowed by the given model. For cost, it may be useful to set this lower than the actual max, as a long conversation or a question in a long public channel can generate a lot of context.
• HIDDEN_PROMPT_URL: A url containing text to provide as the system prompt to the bot. All paragraph tags, whose text does not start with #, will be included as paragraphs at the start of the hidden prompt. The URL will be refetched once every five minutes. A Google Doc that is published to the web works well as a hidden prompt URL.

(lazy requirements for now)

pip install requirements.txt

Copy .env.example to .env and write credentials.

Run dev server:

watchmedo auto-restart python aibot.py

This uses sockets to talk to Slack, so dev bot will be live if you're using live credentials.

Just keep python aibot.py running on the server. See Procfile for running with Dokku.