DogBot 🐶

art by Yookie

/summarize_chat

Summarize recent messages sent in a group chat.

Defaults to a neutral style, but a custom style can be provided:

Custom style

/summarize_chat as a script for a podcast hosted by talking dogs

/summarize_url

Attempts to summarize the main content of a web page.

Defaults to a neutral style, but a custom style can be provided:

Custom style

/summarize_url https://example.com/news_article as though it's being presented as evidence in a court case

EXHIBIT A: E. coli Outbreak Linked to McDonald's Quarter Pounders

SUMMARY OF KEY FINDINGS

1. Outbreak Overview: An E. coli outbreak linked to McDonald's Quarter Pounders has led to at least 49 illnesses across 10 states, including one death.
2. Source of Contamination: A specific ingredient has not been confirmed as the source of the outbreak,

...

/vibe_check

Analyze chat members' moods

• Summit: 💻 📊 🤖 / inquisitive, methodical, redundant
• SomeUser: 😍 ✨ 💖 / enamored, zealous, effusive
• AnotherUser: 😩 📉 😒 / despondent, lethargic, irritable
• ...

/translate french hola mi amigo

Translates the text to requested language, or English by default.

Alternatively, just reply to any message from the chat and type /translate to translate that message.

Supported languages (using suggested model Aya-23):

Arabic, Chinese (simplified & traditional), Czech, Dutch, English, French, German, Greek, Hebrew, Hindi, Indonesian, Italian, Japanese, Korean, Persian, Polish, Portuguese, Romanian, Russian, Spanish, Turkish, Ukrainian, and Vietnamese

/chat_stats

Print statistics about the chat (only knows about stuff that's happened since bot was added to room)

📊 Chat Stats
  • Total Messages: 100
  • Last 2 days: 40 (40%)

🗣 Top Yappers - 2 days
  1. SomeUser / 30 msgs (75%)
  2. Summit / 10 msgs (25%)

⭐ Top Yappers - all time
  1. Summit / 70 msgs (70%)
  2. SomeUser / 30 msgs (30%)

Features & Ideas

• LLM chatroom summarization
  • Aware of reply threads
  • Aware of media presence (photo/video/etc) & captions on media
  • Aware of current date/time and times of chat messages
  • "Vibe check" summary of users' moods
• Translate messages & replied messages between different languages
• Summarize URLs
  • If /summarize_url command is followed by a URL
  • or /summarize_url command is a reply to another msg containing a URL
• Talk to the bot - send a message tagging bot using @itsUsername
  • Aware of current date/time and times of chat messages
• Nightly auto-delete of all chat data > 2 days old
• Automatically transcribe all voice messages sent in the chat & translate to English
• User settings/personalization
  • Allow users to add a very short "bio" to be included with their messages in prompts so LLM functions use correct pronouns etc.
• Jannie features
  • Granular authorization: only admins/mods can execute commands, etc.
  • Customizable old-message-deletion timeframe

Deployment

Designed to be deployed on Heroku with HuggingFace as the LLM API provider, but should be adaptable to anything.

• Handles 8000+ messages/day on 1x basic dyno with the cheapest Postgres ($7/month and $5/month respectively).
• HuggingFace Pro offers an OpenAI-compatible API for $9/month.

Steps

1. Copy the master key encrypting your rails prod credentials config/credentials/production.key and make it available on your server as an environment variable RAILS_MASTER_KEY.
2. You can also set RAILS_SERVE_STATIC_FILES to disabled if you want.
3. Set up nightly data auto-delete:

Auto-delete old data nightly

A rake task rake nightly_data_purge is set up to purge old messages & other data from the DB. This is intended to be run nightly by a Heroku Scheduler task, but you could use any task scheduling system to run it.

Add the free Heroku Scheduler addon to your app and then configure it with:

1. Run every day at 12:00am UTC (or whenever)
2. Run command: rake nightly_data_purge

Local Development

Install

1. Install postgres, ruby, bundler, heroku CLI.
2. bundle install
3. Setup database or whatever.
4. Configure options in rails credentials (example: credentials.sample.yml).

• App expects two credentials files:
  • config/credentials/production.yml.enc (config/credentials/production.key) is for production only
  • config/credentials.yml.enc (config/master.key) is for dev & test

Run local dev env in poll mode (no webhook)

• just run

  aka

• heroku local --procfile=Procfile.dev

Run local dev env in async/webhook mode

Only use this if you want to locally test the webhooks mode used in production for some reason (requires ngrok).

1. Add your ngrok_url and telegram_secret_token to rails development credentials.
2. Start server with rails s.
3. After you're done, run Telegram.bot.delete_webhook in a rails c console to delete the webhook so poll mode works again.

Run linter & tests

• just test

  (aka rubocop and rspec in parallel)