chatGPToot 🦣 -- Mastodon OpenAI Chatbot 🤖

Description

This is a Mastodon chatbot that uses OpenAI to generate responses. The bot can post toots and DALL-E images based on a pre-written prompt or respond to direct mentions.

---

| 🆕 - Updates - 🆕 | 🤖 - Mastodon Mention Commands - 🤖 | ✳️ - Dependencies - ✳️ | 💾 - Installation - 💾 | ⚖️ - License - ⚖️ |

---

Stats

---

Dependencies

Bot

Status Server

Installation

assumes you already have a Mastodon bot account and have your secret token, as well as an OpenAI API key

• Clone the repository: git clone https://github.com/skullzarmy/chatGPToot.git
• npm install to initiate node and install dependencies
• npm run setup will create needed folders
• cp .env.example .env will copy .env.example to .env file
• nano .env or use text editor to edit .env file and add your credentials
• nano config.json or use text editor to edit config.json file
  • IMPORTANT Your bot will toot about all of my blog posts if you do not edit this!
  • I STRONGLY recommend running npm run ingest-feeds after adding a new RSS feed, or else it will toot about each new one it finds. There is no mechanism to modulate number or tempo, it will just spit them out! Ingesting the feed will add all existing items to local file memory, and check against that file for new items in the loop.

.env

MASTODON_ACCESS_TOKEN=GET-FROM-MASTODON
MASTODON_API_URL=YOUR-MASTODON-INSTANCE
OPENAI_KEY=sk-...
MASTODON_ACCOUNT_ID=GET-FROM-MASTODON
MASTODON_ADMIN_ACCOUNT=TO-ENABLE-ADMIN-COMMANDS
MASTODON_ADMIN_ALERT_USERNAME=TO-ALERT-FOR-FEEDBACK
NEWSDATA_API_KEY=SET-UP-A-FREE-KEY-TO-GET-NEWS-DATA
#STATUS_API_TOKEN=GENERATE-NEW-PRIVATE-KEY-TO-AUTH-STATUS-REQUESTS

config.json

{
    "rss_urls": ["http://YOUR-RSS-FEED-URL/rss"],
    ...
}

(top)

Arguments

The following arguments are available for app/chatgptoot.js:

--no-loop - Disables the automatic toot, image, and mention loops. This is useful if you want to run the script manually to test it.

--no-image - Disables the automatic image loop.

--no-toot - Disables the automatic toot loop.

--no-mention - Disables the automatic mention loop.

--no-rss - Disables the RSS checker loop.

--toot-now - Generates a toot and posts it immediately.

--image-now - Generates an image prompt and posts it immediately.

(top)

NPM Scripts

npm run bot will run the bot in nohup a loop, tooting, mentioning, and posting images.

npm run bot-tail will run the bot as above and stream the logs to your terminal (by tailing the log file)

npm run bot-mention will run the bot in a loop, mentioning but not tooting or posting images.

npm run bot-image will run the bot in a loop, posting images but not tooting or mentioning.

npm run bot-toot will run the bot in a loop, tooting but not mentioning or posting images.

npm run single-toot will run the bot once, tooting but not mentioning or posting images.

npm run single-image will run the bot once, posting images but not tooting or mentioning.

npm run single-mention will run the bot once, mentioning but not tooting or posting images.

npm run tail-logs will stream the new bot logs to your terminal.

npm run usage will generate a usage report located in reports/

npm run status-server will start an express server on port 3000 & ngrok tunnel. Config in app/status_http_server.js

• Need to configure ngrok config auth-token ... from ngrok.
• I plan to make this more modular and move config to config.js eventually

npm run ingest-feeds will loop through your RSS feeds and ingest the existing items into a local file so they are not considered 'new'.

• HIGHLY RECOMMENDED THE FIRST TIME YOU ADD A NEW RSS FEED

(top)

Mention Commands

The bot supports the following commands when responding to a mention:

NOTE: Bot will now respect and maintain your visibility settings. If you DM, it will DM back.

command must be at the beginning of the mention (after the @mention)

• "{no command just text}" - Bot will respond to direct mentions. If no command is used, bot will use your mention text and generate a chat completion for reply.
• "//news//" – For now this only checks for a few articles with keywords "gpt OR ai OR llm OR openai OR copilot OR midjourney" in Science or Technology categories. It will then read and summarize them for you. I plan to expand this to take prompts for search criteria but the news API costs $$. Feel free to buy me a beer to help support the development!
• "//feedback// {feedback}" - Logs user feedback for review
• "//beta//" or "//beta-application//" - Posts the application link to the user
• "//help//" or "//commands//" - Posts a list of commands in reply

Following Only (bot account must follow user sending command)

• "//img// {prompt}" or "//image// {prompt}" - Treats as an image prompt, generates a DALL-E image, and posts it in reply
• "//img-asst// {prompt}" or "//image-assist// {prompt}" - Asks GPT-3 to revise your idea into a new image prompt, then sends for DALL-E image generation and returns image and prompt in reply.

(top)

Admin Only

• "//toot-now// {prompt}" - Posts a new toot to main bot account using given prompt
• "//image-now// {prompt}" - Posts a new image to main bot account using given prompt
• "//status//" - Replies with a status message.

Example

@chatGPToot //image// a cat eating a taco

(top)

Usage Logging Utilities

By default, the bot will log all usage to logs/usage_logs.json

npm run usage will print and save a usage summary to reports/

(top)

RSS Feed Subscription

Bot can now subscribe to RSS feeds, track previously known items, and generate toots about newly found items in the feed.

Modify the config.json file and add any number of RSS feeds to follow. Be careful! It will auto-post all newly found items without any care in the world about spamming everyone!

GPT-4 Suggested Improvements

[COMPLETED]

To Do

• Just realized that for some reason user ID will change from time in the Mastodon responses, leading to a single user logging under multiple IDs. I will have to probably just log under username instead to avoid this. For now, usage logging is somewhat broken.
• Implement redis for persistent bottleneck rate limiting.
• Consider persistent memory, DB, etc. for long-term storage of settings.
• Subscribe feature - Users can subscribe to various toot types to be sent on a dynamic interval. e.g. Make me an image of a penguin every morning.
  • Should have persistent memory.
• Implement fs/promise in chatgptoot.js (download image broke).
• Consider restricting news injection in toot generation to specific times so toots are not always news related.

(top)

Updates

• 2023-04-13
  • Removed dependency for deprecated requests package. This required refactoring the node-mastodon package, which has been abandoned. I removed the dependency on this package and added modules/tusk/ which is largely the same code, refactored to use axios and utilize the vanilla js Promise functionality. Dependabot can leave me alone about it now.
  • Setup an express server that serves a status JSON object. It checks my blog site and the bot node process and returns current status messages for both.
  • Added RSS feed subscription. Define RSS URLs in config.json
  • Bot account was suspended, so I moved to a new account on @botsin.space. Bot is renamed Mr. Roboto (@mrroboto). I have decided not to change the repo name. Botsin.space does not enable the trending/tags API endpoint, which was used in addContext(), so I commented out the getTrendingTags() call but left it in place for running on instances that do support it. I may go to the trouble of setting up a separate account on another instance just to get trending tag data, but I highly doubt it.
  • Instead of trending tags, I added a news checker app/news_handler.js which will return the latest news. addContext() is making use of this now to enrich the chat completion's context. I have noticed it is more heavily weighing the output than I would like, so I may consider putting some trigger variable into the function invocation to control weather I want the news added or not. Then set some mechanism for deciding which of the scheduled toots will be news related. Maybe I will setup a whole separate loop for it.
• 2023-04-16
  • Cleaned up the README.md. Added npm run bot-tail command and added docs for the server and ingest commands.
  • Slava Ukraini
• 2023-04-17
  • Added alt text to image posts for a11y
• 2023-04-18
  • Moved modules/tusk to new npm package at https://www.npmjs.com/package/tusk-mastodon and moved to npm install.
• 2023-04-21
  • Moved timing of loops to config.json 'intervals[]'
  • Modified News disclaimer. Trying to get it to parse out the ads that come in on the free News API and not act as a marketing bot for large corps that pay for news distribution.
• 2023-04-22
  • Improved usage logging - now logs under username rather than ID number.
  • Changes to support opening up some commands to all Mastodon accounts. Moving toward the potential for fully opening it up.
    • Modified the ratelimiter from 50 per day to 5 per hour per account, or a total of 120 per day if all tokens are used every hour. Obviously a bit dangerous as someone could easily run multiple accounts, but I am planning to keep a close eye on usage, and I have kept image gen to followed only.
    • Modified processMention() to pass the isFollower variable into the handler functions where needed. Removed the main IF isFollower check.
    • Modified various handler functions to check isFollower as needed and moved the denial responses there.
  • Big change to the '//news//' command. Will now loop each article, get a single summary, and stitch it all into a summarized response.
  • Bot now respects and responds in the visibility setting that was used in the mention. Will no longer reply publicly to a DM or unlisted toot.
    • Various modifications to reply formatting to ensure @s are used (especially in multi-toot replies) to ensure DM messages can all be seen by the intended recipient.
• 2023-05-16
  • Added JSDocs documentation to all pertinent files
  • Misc housekeeping & code cleanup
  • In preparation for GPT-4, the model used is now configured in config.json rather than hardcoded.
  • Improvements to postToot() to better handle multi-toots and avoid race condition where secondary toots were not 'in reply to' the thread properly.

(top)

---

---

License

This project is released under the MIT License. You are free to use, modify, and distribute the source code, subject to the terms and conditions of the license.

The MIT License is a permissive open-source software license that allows you to use this project for any purpose, including commercial use. By using this project, you agree to retain the original copyright notice and the full license text in all copies or substantial portions of the software.

We encourage collaboration and contribution to the project. Feel free to fork, modify, and share your improvements with the community. Our goal is to make this software as useful and accessible as possible, and your contributions will help us achieve that.

Please note that this project is provided "as is" without any warranty or liability. The authors are not responsible for any consequences that may arise from the use of this software.

(top)