Unofficial API in Python

Implements direct HTTP client access to ChatGPT's web interface by circumventing Cloudflare protection through TLS-based request spoofing and session management. The V1 API constructs authenticated requests that mimic browser behavior, handling cookie persistence, CSRF tokens, and Cloudflare challenge responses to maintain stateful conversations without relying on OpenAI's official API endpoints. This approach enables free access to ChatGPT models by reusing existing web session credentials.

Provides a structured Python wrapper around OpenAI's official ChatGPT API (gpt-3.5-turbo, gpt-4) with built-in conversation history management, automatic context truncation, and streaming response handling. The V3 API maintains conversation state in memory or via external storage, automatically manages token limits by truncating older messages, and abstracts away raw API request/response formatting. This enables developers to build multi-turn conversational applications without manually managing conversation context or token counting.

Implements conversation threading using message IDs and parent IDs to track conversation structure and enable branching conversations. Each message has a unique ID and references a parent message ID, allowing the system to reconstruct conversation trees and support multiple conversation branches from a single parent. This enables features like conversation forking, editing previous messages, and exploring alternative conversation paths. The system tracks conversation IDs for grouping related messages.

Supports configurable HTTP/HTTPS proxies and custom network settings for accessing ChatGPT in restricted network environments (corporate firewalls, VPNs, etc.). The system accepts proxy URLs in configuration, passes them to the underlying HTTP client (requests for sync, aiohttp for async), and handles proxy authentication. This enables the library to work in environments where direct internet access is blocked or monitored. Both V1 and V3 APIs support proxy configuration.

Implements flexible authentication for the V1 reverse-engineered API supporting both email/password login and direct access token injection. The system handles OpenAI's authentication flow including optional captcha solving via external services (2captcha, hcaptcha), session token refresh, and credential validation. For V3, it accepts OpenAI API keys directly. This abstraction allows developers to choose authentication method based on their security posture and automation requirements.

Implements a plugin architecture (V1 only) that allows ChatGPT to invoke external tools and services during conversation. The system maintains a plugin registry loaded from configuration, detects when the model requests plugin execution, and routes requests to appropriate plugin handlers. Plugins can be web APIs, local functions, or external services — the framework handles serialization, error handling, and response injection back into the conversation context. This enables ChatGPT to perform actions beyond text generation (web search, calculations, database queries).

Implements streaming response processing for both V1 and V3 APIs, delivering model output tokens in real-time as they are generated rather than waiting for complete response. The system parses server-sent events (SSE) or chunked HTTP responses, extracts individual tokens, and yields them to the caller. This enables responsive user interfaces with progressive text rendering, reduced perceived latency, and better user experience in web/mobile applications. Supports both synchronous iteration and asynchronous streaming.

Provides fully asynchronous Python interfaces (using asyncio) for both V1 and V3 APIs, enabling concurrent ChatGPT requests without blocking. The implementation uses async/await patterns, aiohttp for HTTP requests, and async generators for streaming responses. This allows developers to build high-concurrency applications that can handle multiple conversations simultaneously without thread overhead. Both APIs expose async variants of all core methods.

Implements conversation state tracking and automatic context management for multi-turn interactions. The system maintains message history (user and assistant messages), tracks token counts per message, and automatically truncates older messages when approaching token limits. This prevents exceeding model context windows while preserving recent conversation context. Supports both in-memory storage and external persistence hooks. The V3 API includes token counting logic to estimate message costs before sending to OpenAI.

Provides a full-featured CLI application (via __main__.py) for interactive ChatGPT conversations with configuration file support, command history, and model selection. Users can start conversations, switch models, manage conversation history, and configure authentication through command-line arguments or YAML config files. The CLI supports both V1 and V3 APIs, streaming output, and readline-based input with history. This enables non-programmers and developers to interact with ChatGPT without writing code.

Implements flexible configuration system supporting YAML config files and environment variables for API credentials, model selection, proxy settings, and behavior tuning. The system loads configuration hierarchically (defaults → config file → environment variables), allowing developers to manage different configurations for dev/staging/production without code changes. Supports per-API-version configuration with validation and sensible defaults. This enables secure credential management and environment-specific behavior.

Defines custom exception types (in typings.py) for different failure modes (authentication errors, rate limits, network errors, API errors) and implements error handling throughout both V1 and V3 APIs. The system catches low-level HTTP errors, parses API error responses, and raises typed exceptions that developers can catch and handle specifically. This enables graceful error handling, retry logic, and user-friendly error messages. Supports both synchronous and asynchronous error propagation.