chat-relay/docs/provider-chatgpt.md

ChatGPT Provider Architecture

This document describes the implementation details of the ChatGptProvider class used to automate and capture interactions with chatgpt.com.

---

🧩 Overview

ChatGptProvider enables message injection, UI automation, and response capture from ChatGPT via DOM or Chrome Debugger methods. It uses retry logic for robust message delivery and streaming capture for response chunks.

---

⚙️ Configurable Parameters

this.captureMethod = "debugger"; // or "dom"
this.debuggerUrlPattern = "*chatgpt.com/backend-api/conversation*";
this.includeThinkingInMessage = true;

These control the response source (network or UI) and message formatting behavior.

---

📌 DOM Elements

• Input: #prompt-textarea
• Send button: #composer-submit-button
• Response area: .message-bubble .text-content
• Loading spinner: .loading-spinner
• Fallback DOM: .message-container .response-text

---

🔄 Lifecycle

1. Initialization

• Sets up DOM selectors and state containers
• Initializes request tracking maps and debug logs

2. Sending Messages

• sendChatMessage():
  • Finds input + button, injects message
  • Retries on failure (up to 5 times)
  • Waits between attempts and checks element readiness

3. Capturing Responses

A. Debugger Mode

• Callback registration via initiateResponseCapture()
• Processes data in handleDebuggerData()
• Uses accumulator map for text chunk assembly and parseDebuggerResponse() to interpret SSE stream format

B. DOM Mode

• Starts DOM observer loop via _startDOMMonitoring()
• Stops when _isResponseStillGeneratingDOM() returns false

---

🛡️ Error Handling

• _reportSendError() reports issues back to callback
• Handles:
  • Missing input or button
  • Disabled controls
  • Empty raw debugger data
  • Unparseable or non-relevant JSON payloads

---

🧠 Streaming SSE Parse Logic

sequenceDiagram
    participant Debugger
    participant Provider
    participant ContentScript

    Debugger-->>Provider: data: { chunk }
    Provider->>Provider: parseDebuggerResponse()
    Provider-->>ContentScript: Accumulated text + isFinal

---

✅ Summary

The ChatGPT provider is designed for robust interaction with ChatGPT's UI or network. It supports retries, error recovery, and chunked response reconstruction, ensuring high reliability and compatibility across updates to the site's frontend.