revChatGPT: Python Library for ChatGPT API Access
revChatGPT bundles two different things behind one name: a reverse-engineered wrapper around ChatGPT's web session (V1) and a thin client for OpenAI's official paid Chat API (V3). V1 is the interesting half — free access through your existing ChatGPT login, routed via a Cloudflare bypass proxy — but it rides on a scraping trick OpenAI could break without notice. If you want something stable to build on, use V3 and skip the V1 headache.
What is revChatGPT?
revChatGPT is a Python package, installed via pip, that gives programmatic access to ChatGPT through two modules. V1 logs into your existing ChatGPT account with an email/password or an access token and routes requests through an open-source Cloudflare bypass proxy. V3 wraps OpenAI's official, paid Chat Completions API and needs only an API key. Both expose a Chatbot class with an ask() method plus a command-line interface.
Key features of revChatGPT
- ✓Two auth paths for V1: email/password login (not supported for Google or Microsoft-linked accounts) or a manually copied access_token from chat.openai.com/api/auth/session
- ✓Config file at $HOME/.config/revChatGPT/config.json holds credentials, model choice, conversation_id/parent_id, and plugin settings instead of hardcoding secrets in your script
- ✓Model switching in V1 config across gpt-4, gpt-4-browsing, gpt-4-plugins, and text-davinci-002-render-sha
- ✓Plugin support in V1 — get_plugins() lists installed plugins and install_plugin(plugin_id=...) installs new ones from code, including unverified plugin domains if you opt in
- ✓V3 exposes ask() and ask_stream() against OpenAI's official API, with CLI flags for temperature, top_p, reply_count, base_prompt, and truncate_limit
- ✓Command-line interface for both versions with multi-line input, arrow-key history editing, and autocomplete against previous prompts
- ✓CHATGPT_BASE_URL environment variable points V1 at your own deployed proxy instead of the maintainer's
Using revChatGPT in your Python code
For V1, streaming a reply looks like: `from revChatGPT.V1 import Chatbot` then `chatbot = Chatbot(config={"access_token": "<token>"})`, and you iterate `chatbot.ask("Hello world")`, reading the growing `data["message"]` string as chunks arrive. For a single blocking result, loop the same way and just keep the last `data["message"]`. V3 is simpler. `from revChatGPT.V3 import Chatbot` then `chatbot = Chatbot(api_key="<api_key>")`, and you call `chatbot.ask("Hello world")` for a full response or `chatbot.ask_stream("Hello world")` to print tokens as they arrive. Both versions also run standalone from the terminal — `python3 -m revChatGPT.V1` or `python3 -m revChatGPT.V3 --api_key <api_key>` — with multi-line input finished by pressing Esc then Enter.
Installing and configuring revChatGPT
Install with `python -m pip install --upgrade revChatGPT`. The README lists Python 3.9 as the minimum and recommends 3.11+. For V1, you need a ChatGPT account, then either your email/password or an access_token copied from https://chat.openai.com/api/auth/session; save whichever you use, plus optional settings like model or proxy, into $HOME/.config/revChatGPT/config.json. On Windows the README says you must set a HOME environment variable yourself, or the library can't find that config file. For V3 you only need an OpenAI API key from platform.openai.com/account/api-keys — no config file required.
Strengths
- ✓One Chatbot class covers both the free (V1) and official paid (V3) route, so switching between them is a one-line import change
- ✓The config file keeps credentials and model choice out of your source code, and the access_token path sidesteps password login entirely
- ✓Streaming works out of the box in both versions — partial `message` chunks in V1 or `ask_stream()` tokens in V3, instead of waiting on a full response
- ✓The CLI is usable on its own for quick terminal chats without writing any Python
Limitations and risks
- △V1 depends on a Cloudflare bypass proxy the maintainer runs, or one you self-host — a workaround for anti-bot protection, not a supported integration, so it can stop working without warning
- △The README caps V1 at 5 requests per 10 seconds on the proxy and 50 requests per hour per OpenAI account, which rules out anything above hobby-scale traffic
- △Email/password login explicitly does not work for accounts linked through Google or Microsoft, forcing those users onto the access_token flow instead
- △V3's GPT-4 model list is fixed to the specific snapshots named in the README (gpt-4, gpt-4-0314, gpt-4-32k, gpt-4-32k-0314, gpt-4-0613) rather than following OpenAI's newest models automatically
- △The project is explicitly unofficial and unaffiliated with OpenAI, so there's no vendor support line if a request breaks
Alternatives to revChatGPT
Frequently asked questions
revChatGPT accepts an access_token copied from https://chat.openai.com/api/auth/session, saved either in the config dict passed to Chatbot() or in $HOME/.config/revChatGPT/config.json. This path works even for Google or Microsoft-linked accounts, unlike email/password login.
The README sets revChatGPT's V1 proxy limit at 5 requests per 10 seconds, on top of OpenAI's own cap of 50 requests per hour for each ChatGPT account used through the wrapper.
revChatGPT's V1 module supports GPT-4 by setting model to gpt-4, gpt-4-browsing, or gpt-4-plugins in the config, and plugins can be listed with get_plugins() or installed from code with install_plugin(plugin_id=...).
revChatGPT is explicitly an unofficial, reverse-engineered project with no affiliation to OpenAI, so its V1 module can break whenever OpenAI changes its web session or anti-bot protection. The README calls it a personal project, which is reason enough to be cautious about production use of that module.
revChatGPT's README lists Python 3.9 as the minimum supported version, with Python 3.11 or newer recommended for the best experience.
revChatGPT's V1 module reuses your existing ChatGPT web login through a Cloudflare bypass proxy, giving free access tied to whatever plan you already have. V3 instead wraps OpenAI's official, paid Chat API directly, requiring an API key from platform.openai.com rather than a ChatGPT account login.
The problem it solves
Anyone who wants to drive ChatGPT from a script has two annoying options: pay OpenAI for full API access, or hand-roll the browser session handling every time login mechanics change on the web app. revChatGPT's V1 module targets the second option — reusing whatever ChatGPT plan you already have, without an extra API bill, by wrapping the Cloudflare-fronted web session in a Python class. V3 exists for people who'd rather pay OpenAI directly but still want a small, scriptable wrapper instead of hand-written HTTP calls.
Best use cases
- •Prototyping a chatbot against your personal ChatGPT plan without paying for API credits, using V1's access_token path
- •Building a script or bot around OpenAI's official paid API through V3, when you already hold a platform.openai.com API key
- •Testing prompts or plugin behavior from a terminal via the built-in CLI instead of the chat.openai.com browser interface
Who should try it — and who should skip
Reach for revChatGPT's V3 module if you already pay for OpenAI's API and just want a lightweight Python wrapper with a CLI on top — that part is stable and unremarkable in a good way. Reach for V1 only if you specifically need to script your existing ChatGPT web login for free and can tolerate a proxy dependency that OpenAI's own anti-bot changes could break at any time. Skip the whole project if you need guaranteed uptime for a production system: an unofficial reverse-engineered login flow is not something to build a paying product on.
Related repositories
Is ChatGPT worth your time?
ChatGPT, Claude and Perplexity can all read this page. Ask one of them what it makes of ChatGPT.
