telegram-phd-bot

A Telegram bot that acts as an AI-powered research assistant, leveraging OpenAI's GPT, Perplexity and DALL-E models.

#telegramPhdBot

#aiogram==3.2.0

#ChatGPT

#OpenAI

#Perplexity

---

Table of Contents generated with DocToc

• telegram-phd-bot
• Feature
  • AI Integration
  • Image Generation
  • Access Level to Features
  • Chat Management
  • Administrative Feature
  • Contributor Feature
  • Jobs
  • Available Commands
• Getting Started
  • Run Bot Backend
    • Run PhD Task Once
  • Start Use Bot
• Usage Notes
• TODO
• Develop

---

Feature

AI Integration

• Supports multiple AI backends:
  • OpenAI (GPT models)
  • Perplexity
• Automatically responds to messages based on triggers:
  • Text length > 350 characters
  • Messages ending with "..." or ":"
  • Direct mentions (e.g. @MiptPhDBot, I could not be sure that SUSY is not exist anymore)
  • Replies to bot messages
  • Messages containing question marks
• Maintains conversation context for natural dialogue
• Configurable discussion modes:
  • Switch between Perplexity and OpenAI APIs for prompts
  • Switch mention only mode (bot only responds to mentions/replies vs all triggers)

Under the hood it uses completion model and chatGPT as chat completion model. The last one is chosen only when there is a dialog context exists, i.e. it is possible to get previous context (message has replay_to and this source message is in the redis cache).

Image Generation

• Creates images using DALL-E based on text prompts
• Supports direct commands and replies
• Automatically refines image prompts for academic context

Access Level to Features

1. Priority Chats

   • Full access to AI features
   • Automatic response to triggers
   • Uses bot's primary OpenAI token

2. Contributor

   • Users can add their own OpenAI tokens
   • Personalized AI features in specified chats for only to be used by the contributor
   • Token management through /add_openai_token command

3. Superadmin

   • Access to AI features when iteract with bot in any chats
   • Broadcast messages (with media) to all bot aknowledged chats, except excluded ones
   • View usage statistics
   • Chat statistics and anonymised monitoring
   • Token usage tracking
   • Access additional administrative commands

4. Basic

   • Chat ID lookup
   • Help commands
   • Basic message responses (echo)

Chat Management

• Automatic chat tracking
• Welcome messages for new members
• Message caching for context (TODO: do not cache messages where you do not have possiblity to use AI features)
• Support for both private and group chats

Administrative Feature

Via env it is possible to configure:

• Configure chat exclusions for PhD periodic job and broadcast messages
• Configure priorty chats
• Configure superadmins

Contributor Feature

For priorty chats it uses a special TokenApiRequestManager request engine with relay on main token and tokens of contributors. The idea was to add possibility to contribute free trial tokens to the bot to unlock unlimitted power of AI in several prioritised chats. Though, today idea is kind of dead, because there are no free trial tokens, and also we used to use our own private tokens nowadys as them not costs to much.

Jobs

• send work result via cronjob to recently active chats & priority chats according to bot_chat_messages_cache. To exclude you should be in [phd work excluded chats, i.e. TG_PHD_WORK_EXCLUDE_CHATS env variable].

Available Commands

• /help - View available commands
• /switch_discussion_mode - Toggle between Perplexity and OpenAI backends [available to everyone]
• /switch_direct_iteration_only - Toggle whether bot responds only to direct mentions/replies or all triggers [available to everyone]
• ...

Getting Started

Run Bot Backend

Prepare .env as in .example.env and start with docker compose orchestrator.

full env that is used you could find in bot settings.py.

When env is ready, you could start bot backend with:

docker-compose up

Run PhD Task Once

Optionally, (e.g. for debugging) you could run PhD task once:

docker-compose run bot --phd-work-notification-run-once

Start Use Bot

1. Add the bot to your chat
2. Use /help to see available commands
3. For enhanced features:
   • Add your chats to priority chat list in env, or
   • Add your OpenAI token via /add_openai_token

Usage Notes

• OpenAI features require either:
  • Being in a priority chat
  • Having contributor status with a valid OpenAI token
  • Superadmin
• Image generation requires additional OpenAI credits
• Message context is maintained for natural conversations. You could control depth of context fetching by according env (check env section).
• All tokens are stored securely using encryption

TODO

• store tokens with ciphering
• reuse fetched data in handlers from filters
• notify user when token does not work
• add request admin right on join
• add superadmin stats fetch
• broadcast message from superadmin (ignore chats?)
• TODO: aiogram.exceptions.TelegramBadRequest: Telegram server says - Bad Request: message is too long
• add service task to delete all keys with expired updated_chat_ttl.
• manage ai client prompt settings via admin commands
• bot could send images, stickers, but what context to store? it could store meta context probably.
• code should be reorginsed like /completions/{commands, message handlers, etc}
• redis could use json.dumps for messages: (# TODO: could be optimised: use json.dumps for messages.)
• bot could be restructured {ai/{commands, messages, etc}, other_apps}
• add black or linter
• add exception logging sending to telegram admin chat

Develop

To Develop you may use the same docker compose, merely do not forget to rebuild always after changes, e.g. docker-compose up --build. Or write your own docker-compose or override compose with volume mounting:

...
volumes:
  - ./bot/src:/opt
...