xtekky/gpt4free

suggest websites to reverse

xtekky opened this issue ยท 109 comments

xtekky commented

issue to suggest websites to reverse
sites to look out for:

  • no cloudflare ( can have security header or cloudflare tls/ssl block but not clearance)
  • gpt 4 / internet
  • can pass in conversation (more control)
  • stream response
  • for ex: sqlchat.ai / t3nsor.com

current ones:

suggested by @localuser-isback

suggested by @phamxtien

suggested by @kwekewk

@localuser-isback

all sites in https://github.com/gitzboy/testremote

xtekky commented

@localuser-isback could you contact me in telegram or discord for the cloudflare bypass ? I have already done some reverse engineering on it and we could work together

xtekky commented

@localuser-isback tekky#1810

DanielShemesh commented

https://beta.pickaxeproject.com/
a great site that now supports creating personalized Chatbots!
It is limited I believe, but you can automate the process of signup and signin.

hemangjoshi37a commented

@xtekky,

I've discovered a few more websites that might be of interest for this project. Here's a list of the sites I've found:

https://ai-assistant.netlify.app

Additionally, I've developed a Python script that can help you automate the process of extracting API endpoints from these websites. Please find the script below:

import requests
from bs4 import BeautifulSoup

websites = [
    "https://ai-assistant.netlify.app"
]

for website in websites:
    response = requests.get(website)
    soup = BeautifulSoup(response.content, 'html.parser')
    for link in soup.find_all('a'):
        href = link.get('href')
        if href.startswith('/api') or href.startswith('http'):
            print(f"Found API endpoint for {website}: {href}")

Feel free to use this script to gather more API endpoints for the project. If you have any questions or need further assistance, please don't hesitate to ask.

Best regards.

  • ๐Ÿ‘2
wingeva1986 commented

https://www.perplexity.ai (better than gpt-3.5)

  • ๐Ÿ‘3
playing4living commented

Im going to be releasing an API that can bypass cloudflare and give you a working cf_clearance token so we can use it to actually send requests to cloudflare protected sites ๐Ÿ‘

It's awesome! Can't wait to see that API!

  • ๐Ÿ‘1
spacewalkingninja commented

Im going to be releasing an API that can bypass cloudflare and give you a working cf_clearance token so we can use it to actually send requests to cloudflare protected sites ๐Ÿ‘

It's awesome! Can't wait to see that API!

I actually have it working right now and have tested with xtekky and it does work! just planning to get it on a server & domain!

Can you upload it to github pls!
Thanks a lot.

Malte0621 commented

How about https://chat.openai.com/ ๐Ÿค”
(I managed to make a python cli for it lol.. But I don't use it anymore .-.)

(Edit: Someone told me that cloudflare might have blocked datacenter IPs from using it, so ig its maybe useless.. Oh well, I pay for the API nowdays .-.)

Aze1337 commented

I was wondering for the Cloudflare version, can you not just open a web browser which completes the JavaScript challenge & save the cookies for further http requests ?

Aze1337 commented

I'd also like to add that you might not want to abuse those services because someone is paying for what you get as "free requests", I would understand that if you are really poor and you want to test a business idea or whatever, but dont abuse it for a long time because someone will pay the bill at the end of the day.

xtekky commented

I don't use this repo myself, and the terms which you agree to when using this repo say that I am not liable for any action made by users or this repo.

Which means, I costed nothing to these sites

  • ๐Ÿ‘€1
tiagorangel2011 commented

Here are some more:

upintheairsheep commented

Snapchat AI

  • ๐Ÿ‘1
Quynh04 commented
hemangjoshi37a commented

Is the GPT4 model the best according to you guys or is there any even better model for resoning and coding? is llama or alpaca better than GPT4 considering the reasoning capabilities? If anyone has done such comparison please let me know. I want to use it for pregramming and coding help.

SylveonDeko commented

Why are there no c# wrappers for any of these???

tiagorangel2011 commented
zahir2000 commented

Is the GPT4 model the best according to you guys or is there any even better model for resoning and coding? is llama or alpaca better than GPT4 considering the reasoning capabilities? If anyone has done such comparison please let me know. I want to use it for pregramming and coding help.

The open source models have not yet reached the level of GPT4. Recently WizardVicunaLM showed unofficial competing results against gpt3.5 and MPT with a whopping 65K+ token generation

HirCoir commented

issue to suggest websites to reverse sites to look out for:

  • no cloudflare ( can have security header or cloudflare tls/ssl block but not clearance)
  • gpt 4 / internet
  • can pass in conversation (more control)
  • stream response
  • for ex: sqlchat.ai / t3nsor.com

current ones:

suggested by @localuser-isback

suggested by @phamxtien

suggested by @kwekewk

@localuser-isback

all sites in https://github.com/gitzboy/testremote (hell yeah lmao)

Please don't make an OpenAssistant API.
OpenAssistant is a project of a small group of people, which is in development and resources are limited despite being hosted by partners. By creating an API I would be guilty of closing the beta project, I will still contact the OpenAssistant team to take action on it. I remind myself again that this is a project in order to create a free AI model for everyone. You can get access to their models at https://huggingface.co/OpenAssistant

  • ๐Ÿ˜•2
  • โค๏ธ7
huangkezhao commented

** ๅฏไปฅๆฐธไน…ๅ…่ดน็”จ http://aiassist.chat

MemeticGitHubUser commented

when you guys talk about reverse engineering these, do you mean making it so they are a useable API endpoint?

x404xx commented

issue to suggest websites to reverse sites to look out for:

* no cloudflare ( can have security header or cloudflare tls/ssl block but not clearance)

* gpt 4 / internet

* can pass in conversation (more control)

* stream response

* for ex: sqlchat.ai / t3nsor.com

current ones:

* https://play.vercel.ai (gpt4)

* https://personified.me (gpt4)

* https://beta.character.ai

* https://neuroflash.com

* https://openprompt.co

* https://www.sqlchat.ai

* https://www.sincode.ai/

* https://open-assistant.io/ ( no filter )

suggested by @localuser-isback

* https://freegpt.one/

* https://chat.pawan.krd/

* https://gptgo.ai/?hl=en

* https://chatgptproxy.me/#/

* https://essay.notpiracy.org/

* https://www.sincode.ai/

suggested by @phamxtien

* https://www.perplexity.ai/

suggested by @kwekewk

* https://chat.gpt.bz/gpt

@localuser-isback

* http://61.173.37.91:8088/

all sites in https://github.com/gitzboy/testremote (hell yeah lmao)

Done for GPTgo!
https://github.com/x404xx/GPT-Go

Sovenok-Hacker commented

What about Brat AI (Brat GPT) (https://bratgpt.com)?

  • ๐Ÿ‘3
hrizal commented

What about Brat AI (Brat GPT) (https://bratgpt.com)?

chatbot with persona as evil ai ๐Ÿ˜…

technicaljicama commented

Brat uses a token system (header).

Encrypt date.now() using aes and this key: UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9

Sovenok-Hacker commented

Brat uses a token system (header).

Encrypt date.now() using aes and this key: UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9

I think it will be easy to reverse its api, maybe i do it soon

chesnokpeter commented

chatbot.theb.ai

x404xx commented

Brat uses a token system (header).
Encrypt date.now() using aes and this key: UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9

I think it will be easy to reverse its api, maybe i do it soon

Tested! it worked.

now = datetime.datetime.now()
timestamp = str(now.timestamp())
secret_key = "UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9"

hash_ = hmac.new(secret_key.encode(), timestamp.encode(), hashlib.sha256).digest()
token = base64.b64encode(hash_).decode()

CiaranYoung commented

I found a website that gives you unlimited access to gpt4 for $9. This site does the reverse:
https://halist.ai/

Sovenok-Hacker commented

Brat uses a token system (header).
Encrypt date.now() using aes and this key: UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9

I think it will be easy to reverse its api, maybe i do it soon

Tested! it worked.

now = datetime.datetime.now() timestamp = str(now.timestamp()) secret_key = "UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9"

hash_ = hmac.new(secret_key.encode(), timestamp.encode(), hashlib.sha256).digest() token = base64.b64encode(hash_).decode()

it is hmac, not aes

Nebelung-Dev commented

Removed ones others had already said

x404xx commented

Brat uses a token system (header).

Encrypt date.now() using aes and this key: UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9

I think it will be easy to reverse its api, maybe i do it soon

Tested! it worked.

now = datetime.datetime.now() timestamp = str(now.timestamp()) secret_key = "UkXp2s5v8y/B?D(G+KbPeShVmYq3t6w9"

hash_ = hmac.new(secret_key.encode(), timestamp.encode(), hashlib.sha256).digest() token = base64.b64encode(hash_).decode()

it is hmac, not aes

Yes, it's HMAC, not AES. But it worked fine. You can use that token to make requests and still get a response.

Nebelung-Dev commented

This is how https://ai.mcbbs.gq/ works. It is in js but it should be similar. You can pass in conversation.

var messages = []

async function sendMessage(content) {
    messages.push({
        "role": "user",
        "content": content
    })
    
    var chat = await fetch("https://ai.mcbbs.gq/api/openai/v1/chat/completions", {
      "headers": {
        "accept": "*/*",
        "accept-language": "en-US,en;q=0.9",
        "content-type": "application/json",
        "sec-ch-ua": "\"Not/A)Brand\";v=\"99\", \"Google Chrome\";v=\"115\", \"Chromium\";v=\"115\"",
        "sec-ch-ua-platform": "\"Windows\"",
      },
      "referrer": "https://ai.mcbbs.gq/",
      "body": JSON.stringify({
        "messages": messages,
        "model": "gpt-3.5-turbo",
        "temperature": 0,
        "presence_penalty": 2
      }),
      "method": "POST"
    })
    
    var reader = chat.body.pipeThrough(new TextDecoderStream()).getReader();
    
    while (true) {
        var { value, done } = await reader.read();
        if (done) break;
        var result = JSON.parse(value).choices[0].message.content
    
        messages.push({
            "role": "assistant",
            "content": result
        })
    
        console.log("Got response: " + result)
    }
}

sendMessage("What is 1+1?")
technicaljicama commented

Some maybe wants to try azubis.store, It uses chatgpt 4 in my understanding, but have not been able to find out how siren_sign works

nsde commented

Hey everyone, I made a simple script to check if a website uses Cloudflare. Hope it helps :)

The CLI is available at:
https://github.com/nsde/is-it-using-cloudflare#is-it-using-cloudflare

  • ๐Ÿ‘2
LopeKinz commented

This is how https://ai.mcbbs.gq/ works. It is in js but it should be similar. You can pass in conversation.

var messages = []

async function sendMessage(content) {
    messages.push({
        "role": "user",
        "content": content
    })
    
    var chat = await fetch("https://ai.mcbbs.gq/api/openai/v1/chat/completions", {
      "headers": {
        "accept": "*/*",
        "accept-language": "en-US,en;q=0.9",
        "content-type": "application/json",
        "sec-ch-ua": "\"Not/A)Brand\";v=\"99\", \"Google Chrome\";v=\"115\", \"Chromium\";v=\"115\"",
        "sec-ch-ua-platform": "\"Windows\"",
      },
      "referrer": "https://ai.mcbbs.gq/",
      "body": JSON.stringify({
        "messages": messages,
        "model": "gpt-3.5-turbo",
        "temperature": 0,
        "presence_penalty": 2
      }),
      "method": "POST"
    })
    
    var reader = chat.body.pipeThrough(new TextDecoderStream()).getReader();
    
    while (true) {
        var { value, done } = await reader.read();
        if (done) break;
        var result = JSON.parse(value).choices[0].message.content
    
        messages.push({
            "role": "assistant",
            "content": result
        })
    
        console.log("Got response: " + result)
    }
}

sendMessage("What is 1+1?")
import requests
messages = []
def sendMessage(content):
    messages.append({
        "role": "user",
        "content": content
    })
    chat = requests.post("https://ai.mcbbs.gq/api/openai/v1/chat/completions", headers={
        "accept": "*/*",
        "accept-language": "en-US,en;q=0.9",
        "content-type": "application/json",
        "sec-ch-ua": "\"Not/A)Brand\";v=\"99\", \"Google Chrome\";v=\"115\", \"Chromium\";v=\"115\"",
        "sec-ch-ua-platform": "\"Windows\"",
    }, json={
        "messages": messages,
        "model": "gpt-3.5-turbo",
        "temperature": 0,
        "presence_penalty": 2
    })
    result = chat.json()["choices"][0]["message"]["content"]
    messages.append({
        "role": "assistant",
        "content": result
    })
    print("Got response: " + result)
sendMessage("What is 1+1?")
bagusindrayana commented

https://b.ai-huan.xyz/ bing alternative, just remove popup and overlay with display:none in inspect element

  • ๐Ÿ‘1
PranavPurwar commented

This one support gpt-4 (16k), gpt 3.5 (multiple modes), sage, claude, chat-bison-001 https://chat.ylokh.xyz Uses cloudfare though, you may get it working with @localuser-isback 's cloudfare bypass

remyweb commented

This one support gpt-4 (16k), gpt 3.5 (multiple modes), sage, claude, chat-bison-001 https://chat.ylokh.xyz Uses cloudfare though, you may get it working with @localuser-isback 's cloudfare bypass

You can't , it uses ChimeraGPT

nacho00112 commented

Someone was able to reverse engine https://powerchat.top? It was too difficult for me

leeeawa commented

Someone was able to reverse engine https://powerchat.top? It was too difficult for me

`import os
import requests
from datetime import datetime
import base64,hashlib,json
from ...typing import sha256, Dict, get_type_hints

url = 'https://powerchat.top/'
model = ['gpt-3.5-turbo']
supports_stream = True
needs_auth = False
working = True

headers = {
'Origin':'https://powerchat.top',
'Referer':'https://powerchat.top/',
'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/110.0.0.0 Safari/537.36',
'Content-Type':'text/plain;charset=UTF-8',
'Version':'1.0'
}

def _create_completion(model: str, messages: list, stream: bool, **kwargs):
current_time = datetime.now()
timestamp_in_seconds = current_time.timestamp()
timestamp_in_milliseconds = int(round(timestamp_in_seconds * 1000))
sign = str(timestamp_in_milliseconds)+':question:contact_me_to_work_together_hello@promptboom.com'
sign = hashlib.sha256(sign.encode('utf-8')).hexdigest()
data = '{"did":"060ca8eaa0625da25d61ae94d4a2cf99","chatList":'+json.dumps(messages)+',"special":{"time":'+str(timestamp_in_milliseconds)+',"sign":"'+sign+'","referer":"https://github.com/","path":"https://powerchat.top/"}}'
data = base64.b64encode(data.encode('utf-8')).decode()
r = requests.post('https://api.powerchat.top/requestPowerChat',json={'data':data},headers=headers,stream=True)
for chunk in r.iter_content():
yield chunk.decode()

params = f'g4f.Providers.{os.path.basename(file)[:-3]} supports: ' +
'(%s)' % ', '.join([f"{name}: {get_type_hints(_create_completion)[name].name}" for name in _create_completion.code.co_varnames[:_create_completion.code.co_argcount]])`

kanaya23 commented

can anyone reverse engineer this please? https://www.sincode.ai/app/marve

Lachine1 commented

Someone was able to reverse engine https://powerchat.top? It was too difficult for me

`import os import requests from datetime import datetime import base64,hashlib,json from ...typing import sha256, Dict, get_type_hints

url = 'https://powerchat.top/' model = ['gpt-3.5-turbo'] supports_stream = True needs_auth = False working = True

headers = { 'Origin':'https://powerchat.top', 'Referer':'https://powerchat.top/', 'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/110.0.0.0 Safari/537.36', 'Content-Type':'text/plain;charset=UTF-8', 'Version':'1.0' }

def _create_completion(model: str, messages: list, stream: bool, **kwargs): current_time = datetime.now() timestamp_in_seconds = current_time.timestamp() timestamp_in_milliseconds = int(round(timestamp_in_seconds * 1000)) sign = str(timestamp_in_milliseconds)+'questioncontact_me_to_work_together_hello@promptboom.com' sign = hashlib.sha256(sign.encode('utf-8')).hexdigest() data = '{"did":"060ca8eaa0625da25d61ae94d4a2cf99","chatList":'+json.dumps(messages)+',"special":{"time":'+str(timestamp_in_milliseconds)+',"sign":"'+sign+'","referer":"https://github.com/","path":"https://powerchat.top/"}}' data = base64.b64encode(data.encode('utf-8')).decode() r = requests.post('https://api.powerchat.top/requestPowerChat',json={'data':data},headers=headers,stream=True) for chunk in r.iter_content(): yield chunk.decode()

params = f'g4f.Providers.{os.path.basename(file)[:-3]} supports: ' + '(%s)' % ', '.join([f"{name}: {get_type_hints(_create_completion)[name].name}" for name in _create_completion.code.co_varnames[:_create_completion.code.co_argcount]])`
@leeeawa can yoi please put the code in a code block

nacho00112 commented
nacho00112 commented

Someone was able to reverse engine https://powerchat.top? It was too difficult for me

`import os import requests from datetime import datetime import base64,hashlib,json from ...typing import sha256, Dict, get_type_hints

url = 'https://powerchat.top/' model = ['gpt-3.5-turbo'] supports_stream = True needs_auth = False working = True

headers = { 'Origin':'https://powerchat.top', 'Referer':'https://powerchat.top/', 'User-Agent':'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/110.0.0.0 Safari/537.36', 'Content-Type':'text/plain;charset=UTF-8', 'Version':'1.0' }

def _create_completion(model: str, messages: list, stream: bool, **kwargs): current_time = datetime.now() timestamp_in_seconds = current_time.timestamp() timestamp_in_milliseconds = int(round(timestamp_in_seconds * 1000)) sign = str(timestamp_in_milliseconds)+'โ“contact_me_to_work_together_hello@promptboom.com' sign = hashlib.sha256(sign.encode('utf-8')).hexdigest() data = '{"did":"060ca8eaa0625da25d61ae94d4a2cf99","chatList":'+json.dumps(messages)+',"special":{"time":'+str(timestamp_in_milliseconds)+',"sign":"'+sign+'","referer":"https://github.com/","path":"https://powerchat.top/"}}' data = base64.b64encode(data.encode('utf-8')).decode() r = requests.post('https://api.powerchat.top/requestPowerChat',json={'data':data},headers=headers,stream=True) for chunk in r.iter_content(): yield chunk.decode()

params = f'g4f.Providers.{os.path.basename(file)[:-3]} supports: ' + '(%s)' % ', '.join([f"{name}: {get_type_hints(_create_completion)[name].name}" for name in _create_completion.code.co_varnames[:_create_completion.code.co_argcount]])`

But os.path.basename(__file__)[:-3] isn't the same as __name__.split(".")[-1]?

nacho00112 commented

Also to remove ".py" os.path.splitext doesn't perfom better?

deezed420 commented

Why are there no c# wrappers for any of these???

c# is kinda a pain in the neck.

yvanlok commented

This one support gpt-4 (16k), gpt 3.5 (multiple modes), sage, claude, chat-bison-001 https://chat.ylokh.xyz Uses cloudfare though, you may get it working with @localuser-isback 's cloudfare bypass

BRO PUT MY WEBSITE IM FAMOUS

  • ๐Ÿ˜„3
PranavPurwar commented

This one support gpt-4 (16k), gpt 3.5 (multiple modes), sage, claude, chat-bison-001 https://chat.ylokh.xyz Uses cloudfare though, you may get it working with @localuser-isback 's cloudfare bypass

BRO PUT MY WEBSITE IM FAMOUS

lmao

yvanlok commented

yeah i use ChimeraGPT

This one support gpt-4 (16k), gpt 3.5 (multiple modes), sage, claude, chat-bison-001 https://chat.ylokh.xyz Uses cloudfare though, you may get it working with @localuser-isback 's cloudfare bypass

BRO PUT MY WEBSITE IM FAMOUS

lmao

yeah i use chimeragpt. no point reverse engineering me because i literally use an API which is based on reverse engineering.

MemeticGitHubUser commented

Ooh, how about Openrouter? Is that possible?

https://openrouter.ai

johnd0e commented

https://www.t3nsor.tech/ is reverse-engineered here: https://github.com/editor-syntax/gpt-free
There is also https://ora.ai/, but outdated.

CiaranYoung commented

magai.ai,copy.ai,input.ai

  • ๐Ÿ‘Ž2
johnd0e commented

magai.ai,copy.ai,input.ai

None of them them free

birdup000 commented

Aria AI by Opera it's built into the browser I wonder if it's possible to reverse engineer. GPT-3.5 I think

DoxrGitHub commented

how about https://www.llama2.ai/ or https://llama.replicate.dev/ (basically the same thing) there is streaming and you can control system prompt as well as temp, token amount, top p, and the llama size, and it doesn't use cloudflare

DoxrGitHub commented

https://chatbotai.one/

there is streaming, no login, I didn't see cf but idk abt that

Mininaut commented

https://gpttalk.ru/

try this, gpt-3.5-turbo
ะธะทะพะฑั€ะฐะถะตะฝะธะต

tiagorangel2011 commented
  • ๐Ÿ‘3
Dark-V commented

@xtekky

no cloudflare ( can have security header or cloudflare tls/ssl block but not clearance)
looks like, no cloudflare
gpt 4 / internet
No gpt4, custom core that can support 200k context (if pro version, so free only 100k?)
can pass in conversation (more control)
Idk
stream response
Possibly yes, not websocket. /api/append_message
Request. Respone data be like:
data: {"completion":" and","stop_reason":null,"model":"claude-2.1","stop":null,"log_id":"361a2eca85bf1aaa19f0f653ef9570276b970e9e5a0fa20745703990428575fd","messageLimit":{"type":"within_limit"}}
data: {"completion":" I","stop_reason":null,"model":"claude-2.1","stop":null,"log_id":"361a2eca85bf1aaa19f0f653ef9570276b970e9e5a0fa20745703990428575fd","messageLimit":{"type":"within_limit"}}

for ex: sqlchat.ai / t3nsor.com
claude.ai/chat

Working and reversed another API on github -
https://github.com/KoushikNavuluri/Claude-API/blob/main/claude-api/claude_api.py

Meshwa428 commented

guys i found another link, this one is by a well known library owner LANGCHAIN

this is a free hosted website of langchain's openGPT's which allows us to use gpt 3.5 for free with custom tools like search and all other stuff

here is the link
https://opengpts-example-vz4y4ooboq-uc.a.run.app/

You can literally find the docs to reverse it here๐Ÿ˜‚๐Ÿ˜‚๐Ÿ˜‚
https://github.com/langchain-ai/opengpts/blob/main/API.md

@hlohaus please take a look

  • ๐Ÿ‘€2
lin-calvin commented

chatglm.cn(a chinese phone number should be provided๏ผŒ I can help with this)

matt7ress commented

ok no one suggested this before me which is weird
https://blackbox.ai/ auth optional, already tried reverse-engineering and its extremely simple

  • ๐Ÿ‘1
Mortezanavidi commented

https://groq.com/
no auth - exteremly fast speed

  • ๐Ÿš€2
  • ๐Ÿ‘€2
artiomborovinskii commented

I find a free claude 3 at yeschat.ai but it have cloudflare protection

  • ๐Ÿ‘4
Sovenok-Hacker commented

https://duckduckgo.com/chat?ia=chat Extremly easy to reverse

  • ๐Ÿ‘2
Ojciec-Tadeusz commented

chat.cyming.top no login, gpt3.5, gemini, cloudflare(?) - I don't remember

  • ๐Ÿ‘€1