Skip to content

Latest commit

 

History

History
104 lines (85 loc) · 4.82 KB

README.eng.md

File metadata and controls

104 lines (85 loc) · 4.82 KB
Raw

chatgpt-telegram-bot

*Personal helper working on base model OpenAI GPT-3.5-turbo and GPT-4, which are available in the Telegram app. It processes messages from users, sends them to the OpenAI server, and sends back the answers. The bot supports systems of balance tokens for every user, which are used for API-request. Also, there are available referral systems and every user can get bonus tokens. Last but not least it is an opportunity to set custom users systems prompt for special answers style. Generated from bot himself

Bot sends users messages using OpenAI API and returns back generated answer. All requests from users are sent to the admin's personal chat with the bot. Admin has full control under the bot using special admin's commands.

There are available two main language models:

  • Basic: gpt-3.5-turbo-1106, context window 16k tokens;
  • Premium: gpt-4-1106-preview, context window 128k tokens.

The dall-e-3 model is used to generate images.

Preparation for work

First of all, users must create and fill a file .env in the directory of the script with their own value, like in the example:

OPENAI_API_KEY=yourapikey  
TELEGRAM_API_KEY=yourbottoken  
ADMIN_CHAT_ID=123456789

In this file:

  1. OPENAI_API_KEY - API-key from OpenAI;
  2. TELEGRAM_API_KEY - API-token from bot in Telegram;
  3. ADMIN_CHAT_ID - Admin's ID in Telegram;

example.env is a template of the .env file, which u can find in the repository. All you need is to replace the value from the file with your value and rename the file.

Launch bot

python main.py

After the first launch in script directory will automatically create file data.json, which contains all necessary data.

After the next launch script will read data from a previously created file. File updating as work progresses.

By default, every new user has 30k tokens, which can be used for messages to the bot (API request).

Using the referrals link to start work with the bot allows you to get an additional bonus of 20k tokens for you and your friend who invited you.

Communication with bot

Communication with the bot works in the format "one question - one answer", but if you reply to the bot`s message, the next answer will contain the context of this message. It allows us to continue a conversation on one subject without having to provide context in every message.

By default bot uses a standard system prompt for every request - "You are a helpful assistant", but users can change the main prompt using the command /prompt.

Come back standard system prompt by default user can use the command /reset_prompt.

Language model by default - gpt-3.5-turbo-1106, but the user can change it using the command /switch_model. For using the premium model gpt-4-1106-preview it is necessary to have a positive balance of tokens.

File data.json structure

{
  "global": {
    "requests": 0,
    "tokens": 0
  },
  "admin_id": {
    "requests": 0,
    "tokens": 0,
    "balance": 777777,
    "last_request": "2021-01-01 00:00:00"
  },
  "user1": {
    "requests": 0,
    "tokens": 0,
    "balance": 30000,
    "name": "John",
    "username": "@johndoe",
    "last_request": "2021-01-01 00:00:00",
    "prompt": "You are Marv - a sarcastic reluctant assistant"
  }
}
  1. "global" - general bot`s statistics about the number of requests and used tokens.
  2. Admin`s statistics: number of requests, used tokens, infinity tokens balance, and the last message date.
  3. Statistics for every user: number of requests, used tokens, actual balance of tokens, name, @username, and the last message date.

The commands for users

/start - the start of work: registration in a system and getting the initial balance of tokens;
/help - list of available commands;
/invite or /ref - getting referral link for inviting new users;
/stats - displaying statistics on the number of requests and the amount of tokens used;
/balance - current balance of common and premium tokens;
/prompt or /p - set your own system prompt;
/reset_prompt - reset system prompt on standard;
/topup - command placeholder for replenishing balance;
/ask_favor - request for free tokens to admin;
/imagine or /img or /i - generating an image using text description;
/switch_model or /switch or /sw - switch active language model;
/pro or /prem or /gpt4 - making a quick request using GPT-4 without switching language model.

The commands for administrators

/data or /d - sending to admin copy of the file data.json or data about the user using his user_id or @username;
/refill or /r - refill user balance;
/block or /ban - block user;
/stop - stops the bot completely;
/announce or /a or /notify - send a message to all or chosen users;
/recent_users or /recent - get a list of active users for last N days.