Written by @xtekky & maintained by @hlohaus

By using this repository or any code related to it, you agree to the legal notice. The author is not responsible for any copies, forks, re-uploads made by other users, or anything else related to GPT4Free. This is the author's only account and repository. To prevent impersonation or irresponsible actions, please comply with the GNU GPL license this Repository uses.

Note

^{Lastet version:
Stats:}

pip install -U g4f

docker pull hlohaus789/g4f

🆕 What's New

• Join our Telegram Channel: t.me/g4f_channel
• Join our Discord Group: discord.gg/XfybzPXPH5
• Explore the g4f Documentation (unfinished): g4f.mintlify.app | Contribute to the docs via: github.com/xtekky/gpt4free-docs

Feedback

You can always leave some feedback here: https://forms.gle/FeWV9RLEedfdkmFN6

To do

As per the survey, here is a list of improvements to come

• golang implementation
• Improve Documentation (on g4f.mintlify.app) & Do video tutorials
• Improve the provider status list & updates
• Tutorials on how to reverse sites to write your own wrapper (PoC only ofc)
• Improve the Bing wrapper. (might write a new wrapper in golang as it is very fast)
• Write a standard provider performance test to improve the stability
• update the repository to include the new openai library syntax (ex: Openai() class)
• Potential support and development of local models
• improve compatibility and error handling

📚 Table of Contents

• 🆕 What's New
• 📚 Table of Contents
• 🛠️ Getting Started
  • Docker container
    • Quick start
  • Use python package
    • Prerequisites
    • Install using pypi
  • Docker for Developers
• 💡 Usage
  • The Web UI
  • The g4f Package
    • ChatCompletion
      • Completion
      • Providers
      • Using Browser
      • Async Support
      • Proxy and Timeout Support
  • Interference openai-proxy API
    • Run interference API from PyPi package
    • Run interference API from repo
• 🚀 Providers and Models
  • GPT-4
  • GPT-3.5
  • Other
  • Models
• 🔗 Related GPT4Free Projects
• 🤝 Contribute
  • Create Provider with AI Tool
  • Create Provider
• 🙌 Contributors
• ©️ Copyright
• ⭐ Star History
• 📄 License

🛠️ Getting Started

Docker container

Quick start:

1. Download and install Docker
2. Pull latest image and run the container:

docker pull hlohaus789/g4f
docker run -p 8080:8080 -p 1337:1337 -p 7900:7900 --shm-size="2g" hlohaus789/g4f:latest

3. Open the included client on: http://localhost:8080/chat/ or set the api base in your client to: http://localhost:1337/v1
4. (Optional) If you need to log in to a provider, you can view the desktop from the container here: http://localhost:7900/?autoconnect=1&resize=scale&password=secret.

Use python package

Prerequisites:

1. Download and install Python (Version 3.10+ is recommended).
2. Install Google Chrome for providers with webdriver

Install using pypi:

Install all supported tools / all used packages:

pip install -U g4f[all]

Install required packages for the OpenaiChat provider:

pip install -U g4f[openai]

Install required packages for the interference api:

pip install -U g4f[api]

Install required packages for the web interface:

pip install -U g4f[gui]

Install required packages for uploading / generating images:

pip install -U g4f[image]

Install required packages for providers with webdriver:

pip install -U g4f[webdriver]

Install required packages for proxy support:

pip install -U aiohttp_socks

or:

1. Clone the GitHub repository:

git clone https://github.com/xtekky/gpt4free.git

2. Navigate to the project directory:

cd gpt4free

3. (Recommended) Create a Python virtual environment: You can follow the Python official documentation for virtual environments.

python3 -m venv venv

4. Activate the virtual environment:
   • On Windows:

   .\venv\Scripts\activate
   

   • On macOS and Linux:

   source venv/bin/activate
   

5. Install minimum requirements:

pip install -r requirements-min.txt

6. Or install all used Python packages from requirements.txt:

pip install -r requirements.txt

7. Create a test.py file in the root folder and start using the repo, further Instructions are below

import g4f
...

Docker for Developers

If you have Docker installed, you can easily set up and run the project without manually installing dependencies.

1. First, ensure you have both Docker and Docker Compose installed.

   • Install Docker
   • Install Docker Compose

2. Clone the GitHub repo:

git clone https://github.com/xtekky/gpt4free.git

3. Navigate to the project directory:

cd gpt4free

4. Build the Docker image:

docker pull selenium/node-chrome
docker-compose build

5. Start the service using Docker Compose:

docker-compose up

Your server will now be running at http://localhost:1337. You can interact with the API or run your tests as you would normally.

To stop the Docker containers, simply run:

docker-compose down

Note

When using Docker, any changes you make to your local files will be reflected in the Docker container thanks to the volume mapping in the docker-compose.yml file. If you add or remove dependencies, however, you'll need to rebuild the Docker image using docker-compose build.

💡 Usage

The Web UI

To start the web interface, type the following codes in the command line.

from g4f.gui import run_gui
run_gui()

The g4f Package

ChatCompletion

import g4f

g4f.debug.logging = True  # Enable debug logging
g4f.debug.version_check = False  # Disable automatic version checking
print(g4f.Provider.Bing.params)  # Print supported args for Bing

# Using automatic a provider for the given model
## Streamed completion
response = g4f.ChatCompletion.create(
    model="gpt-3.5-turbo",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True,
)
for message in response:
    print(message, flush=True, end='')

## Normal response
response = g4f.ChatCompletion.create(
    model=g4f.models.gpt_4,
    messages=[{"role": "user", "content": "Hello"}],
)  # Alternative model setting

print(response)

Completion

import g4f

allowed_models = [
    'code-davinci-002',
    'text-ada-001',
    'text-babbage-001',
    'text-curie-001',
    'text-davinci-002',
    'text-davinci-003'
]

response = g4f.Completion.create(
    model='text-davinci-003',
    prompt='say this is a test'
)

print(response)

Providers

import g4f

# Print all available providers
print([
    provider.__name__
    for provider in g4f.Provider.__providers__
    if provider.working
])

# Execute with a specific provider
response = g4f.ChatCompletion.create(
    model="gpt-3.5-turbo",
    provider=g4f.Provider.Aichat,
    messages=[{"role": "user", "content": "Hello"}],
    stream=True,
)
for message in response:
    print(message)

Cookies / Access Token

For generating images with Bing and for the OpenAi Chat you need cookies or a token from your browser session. From Bing you need the "_U" cookie and from OpenAI you need the "access_token". You can pass the cookies / the access token in the create function or you use the set_cookies setter:

from g4f import set_cookies

set_cookies(".bing.com", {
  "_U": "cookie value"
})
set_cookies("chat.openai.com", {
  "access_token": "token value"
})

from g4f.gui import run_gui
run_gui()

Alternatively, g4f reads the cookies with “browser_cookie3” from your browser or it starts a browser instance with selenium "webdriver" for logging in. If you use the pip package, you have to install “browser_cookie3” or "webdriver" by yourself.

pip install browser_cookie3
pip install g4f[webdriver]

Image Upload & Generation

Image upload and generation are supported by three main providers:

• Bing & Other GPT-4 Providers: Utilizes Microsoft's Image Creator.
• Google Gemini: Available for free accounts with IP addresses outside Europe.
• OpenaiChat with GPT-4: Accessible for users with a Plus subscription.

import g4f

# Setting up the request for image creation
response = g4f.ChatCompletion.create(
    model=g4f.models.default, # Using the default model
    provider=g4f.Provider.Gemini, # Specifying the provider as Gemini
    messages=[{"role": "user", "content": "Create an image like this"}],
    image=open("images/g4f.png", "rb"), # Image input can be a data URI, bytes, PIL Image, or IO object
    image_name="g4f.png" # Optional: specifying the filename
)

# Displaying the response
print(response)

from g4f.image import ImageResponse

# Get image links from response
for chunk in g4f.ChatCompletion.create(
    model=g4f.models.default, # Using the default model
    provider=g4f.Provider.OpenaiChat, # Specifying the provider as OpenaiChat
    messages=[{"role": "user", "content": "Create images with dogs"}],
    access_token="...", # Need a access token from a plus user
    stream=True,
    ignore_stream=True
):
    if isinstance(chunk, ImageResponse):
        print(chunk.images) # Print generated image links
        print(chunk.alt) # Print used prompt for image generation

Using Browser

Some providers using a browser to bypass the bot protection. They using the selenium webdriver to control the browser. The browser settings and the login data are saved in a custom directory. If the headless mode is enabled, the browser windows are loaded invisibly. For performance reasons, it is recommended to reuse the browser instances and close them yourself at the end:

import g4f
from undetected_chromedriver import Chrome, ChromeOptions
from g4f.Provider import (
    Bard,
    Poe,
    AItianhuSpace,
    MyShell,
    PerplexityAi,
)

options = ChromeOptions()
options.add_argument("--incognito");
webdriver = Chrome(options=options, headless=True)
for idx in range(10):
    response = g4f.ChatCompletion.create(
        model=g4f.models.default,
        provider=g4f.Provider.MyShell,
        messages=[{"role": "user", "content": "Suggest me a name."}],
        webdriver=webdriver
    )
    print(f"{idx}:", response)
webdriver.quit()

Async Support

To enhance speed and overall performance, execute providers asynchronously. The total execution time will be determined by the duration of the slowest provider's execution.

import g4f
import asyncio

_providers = [
    g4f.Provider.Aichat,
    g4f.Provider.ChatBase,
    g4f.Provider.Bing,
    g4f.Provider.GptGo,
    g4f.Provider.You,
    g4f.Provider.Yqcloud,
]

async def run_provider(provider: g4f.Provider.BaseProvider):
    try:
        response = await g4f.ChatCompletion.create_async(
            model=g4f.models.default,
            messages=[{"role": "user", "content": "Hello"}],
            provider=provider,
        )
        print(f"{provider.__name__}:", response)
    except Exception as e:
        print(f"{provider.__name__}:", e)
        
async def run_all():
    calls = [
        run_provider(provider) for provider in _providers
    ]
    await asyncio.gather(*calls)

asyncio.run(run_all())

Proxy and Timeout Support

All providers support specifying a proxy and increasing timeout in the create functions.

import g4f

response = g4f.ChatCompletion.create(
    model=g4f.models.default,
    messages=[{"role": "user", "content": "Hello"}],
    proxy="http://host:port",
    # or socks5://user:pass@host:port
    timeout=120,  # in secs
)

print(f"Result:", response)

You can also set a proxy globally via an environment variable:

export G4F_PROXY="http://host:port"

Interference openai-proxy API (Use with openai python package)

Run interference API from PyPi package

from g4f.api import run_api

run_api()

Run interference API from repo

If you want to use the embedding function, you need to get a Hugging Face token. You can get one at Hugging Face Tokens. Make sure your role is set to write. If you have your token, just use it instead of the OpenAI api-key.

Run server:

g4f api

or

python -m g4f.api.run

from openai import OpenAI

client = OpenAI(
    # Set your Hugging Face token as the API key if you use embeddings
    api_key="YOUR_HUGGING_FACE_TOKEN",

    # Set the API base URL if needed, e.g., for a local development environment
    base_url="http://localhost:1337/v1"
)


def main():
    chat_completion = client.chat.completions.create(
        model="gpt-3.5-turbo",
        messages=[{"role": "user", "content": "write a poem about a tree"}],
        stream=True,
    )

    if isinstance(chat_completion, dict):
        # Not streaming
        print(chat_completion.choices[0].message.content)
    else:
        # Streaming
        for token in chat_completion:
            content = token.choices[0].delta.content
            if content is not None:
                print(content, end="", flush=True)


if __name__ == "__main__":
    main()

API usage (POST)

Chat completions

Send the POST request to /v1/chat/completions with body containing the model method. This example uses python with requests library:

import requests
url = "http://localhost:1337/v1/chat/completions"
body = {
    "model": "gpt-3.5-turbo-16k",
    "stream": False,
    "messages": [
        {"role": "assistant", "content": "What can you do?"}
    ]
}
json_response = requests.post(url, json=body).json().get('choices', [])

for choice in json_response:
    print(choice.get('message', {}).get('content', ''))

🚀 Providers and Models

GPT-4

Website	Provider	GPT-3.5	GPT-4	Stream	Status	Auth
bing.com	g4f.Provider.Bing	❌	✔️	✔️		❌
free.chatgpt.org.uk	g4f.Provider.FreeChatgpt	✔️	✔️	✔️		❌
liaobots.site	g4f.Provider.Liaobots	✔️	✔️	✔️		❌
chat.openai.com	g4f.Provider.OpenaiChat	✔️	✔️	✔️		✔️
raycast.com	g4f.Provider.Raycast	✔️	✔️	✔️		✔️
beta.theb.ai	g4f.Provider.Theb	✔️	✔️	✔️		❌
chat.geekgpt.org	g4f.Provider.GeekGpt	✔️	✔️	✔️		❌

GPT-3.5

Website	Provider	GPT-3.5	GPT-4	Stream	Status	Auth
chat3.aiyunos.top	g4f.Provider.AItianhuSpace	✔️	❌	✔️		❌
aichatonline.org	g4f.Provider.AiChatOnline	✔️	❌	✔️		❌
openchat.team	g4f.Provider.Aura	✔️	❌	✔️		❌
chatbase.co	g4f.Provider.ChatBase	✔️	❌	✔️		❌
chatforai.store	g4f.Provider.ChatForAi	✔️	❌	✔️		❌
chatgpt.ai	g4f.Provider.ChatgptAi	✔️	❌	✔️		❌
chat.chatgptdemo.net	g4f.Provider.ChatgptDemo	✔️	❌	✔️		❌
chatgpt-free.cc	g4f.Provider.ChatgptNext	✔️	❌	✔️		❌
chat.3211000.xyz	g4f.Provider.Chatxyz	✔️	❌	✔️		❌
gptalk.net	g4f.Provider.GPTalk	✔️	❌	✔️		❌
geminiprochat.com	g4f.Provider.GeminiProChat	✔️	❌	✔️		❌
gpt6.ai	g4f.Provider.Gpt6	✔️	❌	✔️		❌
gptchatly.com	g4f.Provider.GptChatly	✔️	❌	❌		❌
ai18.gptforlove.com	g4f.Provider.GptForLove	✔️	❌	✔️		❌
gptgo.ai	g4f.Provider.GptGo	✔️	❌	✔️		❌
gpttalk.ru	g4f.Provider.GptTalkRu	✔️	❌	✔️		❌
koala.sh	g4f.Provider.Koala	✔️	❌	✔️		❌
app.myshell.ai	g4f.Provider.MyShell	✔️	❌	✔️		❌
onlinegpt.org	g4f.Provider.OnlineGpt	✔️	❌	✔️		❌
perplexity.ai	g4f.Provider.PerplexityAi	✔️	❌	✔️		❌
poe.com	g4f.Provider.Poe	✔️	❌	✔️		✔️
talkai.info	g4f.Provider.TalkAi	✔️	❌	✔️		❌
you.com	g4f.Provider.You	✔️	❌	✔️		❌
aitianhu.com	g4f.Provider.AItianhu	✔️	❌	✔️		❌
e.aiask.me	g4f.Provider.AiAsk	✔️	❌	✔️		❌
chatgpt.bestim.org	g4f.Provider.Bestim	✔️	❌	✔️		❌
chatanywhere.cn	g4f.Provider.ChatAnywhere	✔️	❌	✔️		❌
chatgpt4online.org	g4f.Provider.Chatgpt4Online	✔️	❌	✔️		❌
chat.chatgptdemo.ai	g4f.Provider.ChatgptDemoAi	✔️	❌	✔️		❌
chatgptfree.ai	g4f.Provider.ChatgptFree	✔️	❌	❌		❌
chatgptlogin.ai	g4f.Provider.ChatgptLogin	✔️	❌	✔️		❌
chatgptx.de	g4f.Provider.ChatgptX	✔️	❌	✔️		❌
chat-shared2.zhile.io	g4f.Provider.FakeGpt	✔️	❌	✔️		❌
freegpts1.aifree.site	g4f.Provider.FreeGpt	✔️	❌	✔️		❌
gptgod.site	g4f.Provider.GptGod	✔️	❌	✔️		❌
hashnode.com	g4f.Provider.Hashnode	✔️	❌	✔️		❌
sdk.vercel.ai	g4f.Provider.Vercel	✔️	❌	✔️		❌
chat.ylokh.xyz	g4f.Provider.Ylokh	✔️	❌	✔️		❌

Other

Website	Provider	GPT-3.5	GPT-4	Stream	Status	Auth
bard.google.com	g4f.Provider.Bard	❌	❌	❌		✔️
deepinfra.com	g4f.Provider.DeepInfra	❌	❌	✔️		❌
gemini.google.com	g4f.Provider.Gemini	❌	❌	❌		✔️
huggingface.co	g4f.Provider.HuggingChat	❌	❌	✔️		❌
llama2.ai	g4f.Provider.Llama2	❌	❌	✔️		❌
labs.perplexity.ai	g4f.Provider.PerplexityLabs	❌	❌	✔️		❌
phind.com	g4f.Provider.Phind	❌	❌	✔️		❌
pi.ai	g4f.Provider.Pi	❌	❌	✔️		❌
theb.ai	g4f.Provider.ThebApi	❌	❌	❌		✔️
open-assistant.io	g4f.Provider.OpenAssistant	❌	❌	✔️		✔️

Models

Model	Base Provider	Provider	Website
gpt-3.5-turbo	OpenAI	5+ Providers	openai.com
gpt-4	OpenAI	2+ Providers	openai.com
gpt-4-turbo	OpenAI	g4f.Provider.Bing	openai.com
Llama-2-7b-chat-hf	Huggingface	2+ Providers	huggingface.co
Llama-2-13b-chat-hf	Huggingface	2+ Providers	huggingface.co
Llama-2-70b-chat-hf	Huggingface	4+ Providers	huggingface.co
CodeLlama-34b-Instruct-hf	Huggingface	3+ Providers	huggingface.co
Mixtral-8x7B-Instruct-v0.1	Huggingface	3+ Providers	huggingface.co
Mistral-7B-Instruct-v0.1	Huggingface	3+ Providers	huggingface.co
dolphin-2.6-mixtral-8x7b	Huggingface	g4f.Provider.DeepInfra	huggingface.co
lzlv_70b_fp16_hf	Huggingface	g4f.Provider.DeepInfra	huggingface.co
airoboros-70b	Huggingface	g4f.Provider.DeepInfra	huggingface.co
airoboros-l2-70b-gpt4-1.4.1	Huggingface	g4f.Provider.DeepInfra	huggingface.co
openchat_3.5	Huggingface	2+ Providers	huggingface.co
gemini	Google	g4f.Provider.Gemini	gemini.google.com
gemini-pro	Google	2+ Providers	gemini.google.com
claude-v2	Anthropic	2+ Providers	anthropic.com
pi	Inflection	g4f.Provider.Pi	inflection.ai

🔗 Related GPT4Free Projects

🎁 Projects	⭐ Stars	📚 Forks	🛎 Issues	📬 Pull requests
gpt4free
gpt4free-ts
Free AI API's & Potential Providers List
ChatGPT-Clone
ChatGpt Discord Bot
Nyx-Bot (Discord)
LangChain gpt4free
ChatGpt Telegram Bot
ChatGpt Line Bot
Action Translate Readme
Langchain Document GPT

🤝 Contribute

Create Provider with AI Tool

Call in your terminal the create_provider.py script:

python etc/tool/create_provider.py

1. Enter your name for the new provider.
2. Copy and paste the cURL command from your browser developer tools.
3. Let the AI ​​create the provider for you.
4. Customize the provider according to your needs.

Create Provider

1. Check out the current list of potential providers, or find your own provider source!
2. Create a new file in g4f/Provider with the name of the Provider
3. Implement a class that extends BaseProvider.

from __future__ import annotations

from ..typing import AsyncResult, Messages
from .base_provider import AsyncGeneratorProvider

class HogeService(AsyncGeneratorProvider):
    url                   = "https://chat-gpt.com"
    working               = True
    supports_gpt_35_turbo = True

    @classmethod
    async def create_async_generator(
        cls,
        model: str,
        messages: Messages,
        proxy: str = None,
        **kwargs
    ) -> AsyncResult:
        yield ""

4. Here, you can adjust the settings, for example, if the website does support streaming, set supports_stream to True...
5. Write code to request the provider in create_async_generator and yield the response, even if it's a one-time response, do not hesitate to look at other providers for inspiration
6. Add the Provider Name in g4f/Provider/__init__.py

from .HogeService import HogeService

__all__ = [
  HogeService,
]

7. You are done !, test the provider by calling it:

import g4f

response = g4f.ChatCompletion.create(model='gpt-3.5-turbo', provider=g4f.Provider.PROVIDERNAME,
                                    messages=[{"role": "user", "content": "test"}], stream=g4f.Provider.PROVIDERNAME.supports_stream)

for message in response:
    print(message, flush=True, end='')

🙌 Contributors

A list of the contributors is available here
The Vercel.py file contains code from vercel-llm-api by @ading2210, which is licensed under the GNU GPL v3
Top 1 Contributor: @hlohaus

©️ Copyright

This program is licensed under the GNU GPL v3

xtekky/gpt4free: Copyright (C) 2023 xtekky

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with this program.  If not, see <https://www.gnu.org/licenses/>.

⭐ Star History

📄 License

This project is licensed under GNU_GPL_v3.0.

(🔼 Back to top)