Automatic interaction with the browser version of ChatGPT (without API) to obtain responses, their parsing

🔸 

Goal:

Create a system that allows for automatic interaction with the browser version of ChatGPT (without API) to obtain responses, parse them, and send them back to a Telegram bot.

🔸 

General logic of operation:

1. The user in Telegram clicks a button (for example, “Start Analysis”).
2. The Telegram bot sends a request to the backend server (Node.js / Python).
3. The server opens the page chat.openai.com via Puppeteer or Playwright, inserts the prompt into the input window, and simulates pressing Enter.
4. GPT generates a response — the browser waits for the entire result to load.
5. The script retrieves page.content() or page.evaluate() to get the HTML code of the page.
6. Then, from this HTML, using CSS selectors or XPath, only the required block (assistant's response) is parsed.
7. The obtained text is processed (cleaning, formatting, shortening).
8. The result is returned to the Telegram bot as a regular message or a message with markup.

🔸 

Technical structure:

🔹 1. Telegram Layer

• Node.js or Python (aiogram/Telegraf)
• Webhook or polling
• Buttons, menus, case states

🔹 2. Automation Layer

• Playwright/Puppeteer (main tool for browser control)
• Authorization in GPT via cookies (session storage)
• The script automatically inserts the prompt, presses Enter, waits for a response

🔹 3. Parsing Layer

• HTML parser (node-html-parser or cheerio)
• Extracting the response block:

const lastTurn = await page.$$('article[data-testid^="conversation-turn-"]');

const messageBlock = await lastTurn[lastTurn.length - 1].$('.markdown.prose');

const text = await messageBlock.textContent();

• Cleaning from unnecessary elements (buttons, indicators, html tags)

🔹 4. Data & Storage Layer

• Airtable / Firebase for storing cases and history
• Can add prompt-dispatcher with templates (.jinja2)

🔹 5. Response Layer

• The result is processed (JSON, Markdown, Telegram markup)
• Sent back to the user in Telegram

🔸 

Implementation features:

• Ability to run multiple browser instances simultaneously (for many users)
• Automatic page reload on error
• Support for logs and execution monitoring (Winston / PM2 / cron-healthcheck)
• Optimization of response time (request queue, asynchronous processing)

🔸 

Expected result:

• A fully autonomous system that:
  • accepts requests from Telegram;
  • executes them in browser GPT;
  • receives responses, parses, and returns to the user;
  • supports many simultaneous requests.
•