New Authentication (with Cloudflare) #69

federico123579 · 2022-12-11T22:33:19Z

galligan · 2022-12-11T23:09:03Z

I'm currently getting this one

lyra-ai · 2022-12-11T23:30:18Z

I get the same message above.

Also can't login with the Chromium browser. Cloudflare seems to be hell bent on blocking.

herman925 · 2022-12-12T01:46:27Z

Same for me. I thoguht it was me having a wrong token but it seems it's CF

sluongng · 2022-12-12T03:56:30Z

Got the same with @galligan above.

Dumping the HTML response from OpenAI and got something like this

which seems to be a ddos protection system.

sluongng · 2022-12-12T04:14:07Z

https://github.com/acheong08/ChatGPT/wiki/Setup it seems like CloudFlare protection was added and cf_clearance needs to be set in request's cookies

GitHub
Setup · acheong08/ChatGPT Wiki
Lightweight package for interacting with ChatGPT's API by OpenAI. Uses reverse engineered official API. - Setup · acheong08/ChatGPT Wiki

Some good examples https://sourcegraph.com/search?q=context:global+repo:%5Egithub%5C.com/acheong08/ChatGPT%24+file:%5Esrc/revChatGPT/revChatGPT%5C.py+cf_clearance&patternType=regexp&sm=1

JoshuaDoes · 2022-12-12T18:49:51Z

I've been working within my own fork to try getting things working, but it appears that just by simply adding cf_clearance as a cookie to the session endpoint won't let me through - I've matched my user agent to Firefox and Edge when trying using either browser, and while my browser can get in regardless, my bot can't seem to with the existing environment. This may need some extra work to try to mimic a valid context for CloudFlare to accept us.

tianlichunhong · 2022-12-13T02:49:23Z

I've been working within my own fork to try getting things working, but it appears that just by simply adding cf_clearance as a cookie to the session endpoint won't let me through - I've matched my user agent to Firefox and Edge when trying using either browser, and while my browser can get in regardless, my bot can't seem to with the existing environment. This may need some extra work to try to mimic a valid context for CloudFlare to accept us.

I think use python undetected_chromedriver is a good way to solve the problem. But exactly how to write the code, I'm not sure.

JoshuaDoes · 2022-12-13T03:39:48Z

I've been working within my own fork to try getting things working, but it appears that just by simply adding cf_clearance as a cookie to the session endpoint won't let me through - I've matched my user agent to Firefox and Edge when trying using either browser, and while my browser can get in regardless, my bot can't seem to with the existing environment. This may need some extra work to try to mimic a valid context for CloudFlare to accept us.

I think use python undetected_chromedriver is a good way to solve the problem. But exactly how to write the code, I'm not sure.

I don't feel that's an efficient approach, you'd be better off writing the rest of the bot in Python too if you take that route. This library is pure Go so far, and the implementation already has an existing method of pulling tokens using Playwright to pop open a web browser and automate pulling it after you've manually logged in with the browser instance. That browser instance could be leveraged to automate pulling the CloudFlare cookies in similar ways to how other clients are doing it (such as the node.js and Python clients), but in the end this still relies on a working implementation of the cf_clearance cookie and anything else CloudFlare may now be fingerprinting and detecting. Which is something those other clients didn't need to worry about because their base HTTP clients just work still.

Even then, without using the web browser to automate pulling the cookies, using a predefined session token cookie and CloudFlare clearance cookie should suffice for the manner I'm using my fork of the ChatGPT package in. The way I'm using the underlying package isn't any different to how the Telegram bot is interacting with it, so once it can actually connect to ChatGPT again with the appropriate environment, then it'll be possible to start automating the pulling of cookies again using the web browser in the same way it already is.

UnGrosLoup · 2022-12-13T07:47:43Z

So, any work around so far?

xinmans · 2022-12-13T08:29:37Z

any solution based on openai API?

sluongng · 2022-12-13T12:40:44Z

any solution based on openai API?

This is what I am exploring on my side as well, just don't have enough time.
https://github.com/fxchen/SlackGPT/blob/main/functions/gpt_request.ts is how 1 Slack employee was able to hook up ChatGPT into Slack. https://twitter.com/frankc/status/1602374528897675264 So I think it should be doable? Not quite sure if the model's name is correct there, need some testing.

GitHub
SlackGPT/gpt_request.ts at main · fxchen/SlackGPT
Use this starter application to explore interacting with OpenAI's ChatGPT inside of Slack. I discovered ChatGPT is way more powerful than expected. It's exciting, sometimes wrong, and delig...

Twitter

LyghtCode · 2022-12-13T17:02:23Z

he uses a generated api key from openai, not the session token from browser cookies. or is there no difference? https://github.com/fxchen/SlackGPT/blob/main/functions/gpt_request.ts

GitHub
SlackGPT/gpt_request.ts at main · fxchen/SlackGPT
Use this starter application to explore interacting with OpenAI's ChatGPT inside of Slack. I discovered ChatGPT is way more powerful than expected. It's exciting, sometimes wrong, and delig...

sluongng · 2022-12-14T10:19:16Z

The different seems to be that the model available via OpenAI official API is only GPT-3.5 https://beta.openai.com/docs/models/overview and there is a lack of "context keeping" when using the API vs ChatGPT.

With ChatGPT, you could ask a chain of questions:

how many countries are there?
how about languages?
how about cpu?

and the context is kept on the 3rd question.

But when I attempt to use the API with text-davinci-003, there is no way to let the API know the context and thus, the response quality is lower. But the upside is that there is no Cloudflare by-pass needed and no need for chromium / browser (Im running the bot in a headless raspberry pi)

OpenAI API
An API for accessing new AI models developed by OpenAI

kaixindeken · 2022-12-16T05:07:27Z

Any work around so far?

pupubird · 2022-12-16T05:49:56Z

Facing the same issue

sluongng · 2022-12-16T18:46:34Z

Ah so the path of using the /completion API gets a lot better. I found out about https://beta.openai.com/playground/p/default-chat?lang=curl and realize that you need to send part of the chat history within the prompt. That would provide the AI needed context to continue the conversation.

However, the prompt size is limited at 4000(input and output inclusive) for free account. It's essentially counted per "token" https://openai.com/api/pricing/. So the richer the context, the longer the input prompt, the more expensive it is to use. With a free account, you probably want to set max_tokens to no more than 1000 tokens and leave 3000 tokens for the output?

OpenAI API
An API for accessing new AI models developed by OpenAI

OpenAI
Pricing
OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity.

Lordyzagat · 2022-12-16T21:32:18Z

tianlichunhong · 2022-12-17T07:30:15Z

Ah so the path of using the /completion API gets a lot better. I found out about https://beta.openai.com/playground/p/default-chat?lang=curl and realize that you need to send part of the chat history within the prompt. That would provide the AI needed context to continue the conversation.

However, the prompt size is limited at 4000(input and output inclusive) for free account. It's essentially counted per "token" https://openai.com/api/pricing/. So the richer the context, the longer the input prompt, the more expensive it is to use. With a free account, you probably want to set max_tokens to no more than 1000 tokens and leave 3000 tokens for the output?

**OpenAI API**An API for accessing new AI models developed by OpenAI

OpenAI**Pricing**OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity.

API usage is too expensive. And only 18$ is free. So we need to find a way to permanently access it for free.

OpenAI API
An API for accessing new AI models developed by OpenAI

OpenAI
Pricing
OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity.

pincombe · 2022-12-18T23:01:42Z

maybe thinking outside the box here but maybe we could use FlareSolverr to get around the cloudflare issue. https://github.com/FlareSolverr/FlareSolverr

If I get a bit of spare time next week I'll look to see if i can offer a pull request to integrate this as an option.

GitHub
GitHub - FlareSolverr/FlareSolverr: Proxy server to bypass Cloudflare protection
Proxy server to bypass Cloudflare protection. Contribute to FlareSolverr/FlareSolverr development by creating an account on GitHub.

TTTq · 2022-12-20T09:47:26Z

Use this head to avoid robot detection

args := []string{"--no-sandbox",
	"--disable-setuid-sandbox",
	"--disable-infobars",
	"--disable-dev-shm-usage",
	"--disable-blink-features=AutomationControlled",
	"--ignore-certificate-errors",
	"--no-first-run",
	"--no-service-autorun",
	"--password-store=basic",
	"--system-developer-mode",
	// the following flags all try to reduce memory
	// "--single-process",
	"--mute-audio",
	"--disable-default-apps",
	"--no-zygote",
	"--disable-accelerated-2d-canvas",
	"--disable-web-security",
	// "--disable-gpu"
	// "--js-flags="--max-old-space-size=1024""
}

permblackshark · 2023-03-25T04:14:15Z

Thanks @acheong08 i can build it now. However one dumb question, where do i put the environmental variables?

May be SET VAR_NAME=
or search Environment Variables in windows search

GchinoD · 2023-03-25T05:14:35Z

so after i have set the environment, i just run the initial from @peacecwz ?

acheong08 · 2023-03-25T05:17:01Z

You change the URL to where you're hosting it and then just run this repo I think

GchinoD · 2023-03-25T05:25:42Z

another dumb question @acheong08 where do i get the token or puid from? is it the chatgpt api token?

acheong08 · 2023-03-25T05:27:40Z

Cookie

GchinoD · 2023-03-25T05:50:12Z

erm @acheong08 where do i change the url? and i followed this repo (https://github.com/peacecwz/chatgpt-telegram) to set up to use chatgpt bot for telegram. For your repo, is it possible?

GitHub
GitHub - peacecwz/chatgpt-telegram: Run your own GPTChat Telegram bot, with a single command!
Run your own GPTChat Telegram bot, with a single command! - GitHub - peacecwz/chatgpt-telegram: Run your own GPTChat Telegram bot, with a single command!

acheong08 · 2023-03-25T06:36:58Z

I see that this repo uses Playwright. You'll need to remove that code and just replace all the chat.openai.com with your bypass server (For example, mine is bypass.churchless.tech)

GchinoD · 2023-03-25T08:02:38Z

@acheong08 i have managed to get it to run, but when i tried the bot on my telegram, i get this error though *Error: Couldn't get access token: unauthorized"

Do u have any idea which access token it is talking about?

acheong08 · 2023-03-25T08:03:23Z

https://chat.openai.com/api/auth/session

GchinoD · 2023-03-25T08:15:18Z

@acheong08 so i used the token above and i go run make.go for your repo, and i get error failed to refresh puid cookie

Can i verify the steps i need to do.

set token
run go from your repo
run go from the tg bot repo

is that correct?

acheong08 · 2023-03-25T08:16:17Z

so i used the token above and i go run make.go for your repo, and i get error failed to refresh puid cookie

Did you set export PUID="..."?

acheong08 · 2023-03-25T08:16:39Z

run go from your repo
run go from the tg bot repo

You need to fork their repos and make some changes

GchinoD · 2023-03-25T08:29:49Z

@acheong08 yup i made the changes already. Is it puid or access token?

acheong08 · 2023-03-25T08:30:37Z

PUID for the proxy and access token for both

GchinoD · 2023-03-25T08:33:49Z

Ah where do i get the PUID?

acheong08 · 2023-03-25T08:34:26Z

Cookies of plus account

acheong08 · 2023-03-25T08:34:34Z

_puid

GchinoD · 2023-03-25T08:40:12Z

Im sorry. I think im still confused. So i would need a plus account for this to work?

acheong08 · 2023-03-25T08:40:54Z

One plus account for many free accounts to use without cloudflare

acheong08 · 2023-03-25T08:41:19Z

You can actually just set the URL to my public proxy https://bypass.churchless.tech/api/conversation

acheong08 · 2023-03-25T08:41:37Z

You don't have to host it yourself. It just has rate limits of 5 requests / 10 seconds

GchinoD · 2023-03-25T08:46:51Z

So with yours i do not need the PUID?

acheong08 · 2023-03-25T08:53:45Z

yes since I'm sharing my own PUID

GchinoD · 2023-03-25T15:14:34Z

@acheong08 ok so do I need to change the url link on the main.go file to this: https://bypass.churchless.tech/api/conversation?

How about headers? those are https://chat.openai.com/chat

And also where can i get the PUID of your public proxy?

Anw, really appreciate alot for this help!

acheong08 · 2023-03-25T15:16:14Z

And also where can i get the PUID of your public proxy?

It injects it automatically. You don't need to worry about it

ok so do I need to change the url link on the main.go file to this:

Yes and remove all the puppeteer stuff unless used for login

How about headers? those are chat.openai.com/chat

Ignore them. Don't need to touch it

GchinoD · 2023-03-25T16:06:44Z

i followed the above @acheong08, i got the 405 Method Not Allowed and failed to refresh puid cookie

acheong08 · 2023-03-25T16:07:41Z

i followed the above @acheong08, i got the 405 Method Not Allowed and failed to refresh puid cookie

I mean you don't need to run my repo. Just modify the code of this repo

GchinoD · 2023-03-25T16:11:35Z

ah, but im still getting the unauthorised error when i run this repo with the edits to your proxy
https://github.com/peacecwz/chatgpt-telegram

GitHub
GitHub - peacecwz/chatgpt-telegram: Run your own GPTChat Telegram bot, with a single command!
Run your own GPTChat Telegram bot, with a single command! - GitHub - peacecwz/chatgpt-telegram: Run your own GPTChat Telegram bot, with a single command!

acheong08 · 2023-03-25T16:39:23Z

Is your access token correct?

GchinoD · 2023-03-25T16:59:00Z

I would think so. I coppird the access token part from the link u gave me.

m1guelpf changed the title ~~Error 403: forbidden - presumably because the new authentication system~~ New Authentication (with Cloudflare) Dec 12, 2022

m1guelpf added the help wanted Extra attention is needed label Dec 12, 2022

m1guelpf pinned this issue Dec 12, 2022

m1guelpf mentioned this issue Dec 12, 2022

Error: Couldn't get access token #71

Closed

pluja mentioned this issue Dec 12, 2022

Getting 403 error skydoves/chatgpt-android#10

Closed

This was referenced Dec 13, 2022

Can´t verify I´m human #74

Closed

Error: Couldn't get access token #73

Closed

This was referenced Dec 14, 2022

Error: Couldn't get access token: failed to decode response: invalid character '<' looking for beginning of value #78

Closed

Error: Couldn't get access token #77

Closed

LyghtCode mentioned this issue Dec 15, 2022

Error: Couldn't get access token: failed to decode response: invalid character '<' looking for beginning of value #81

Open

New Authentication (with Cloudflare) #69

New Authentication (with Cloudflare) #69

Comments

federico123579 commented Dec 11, 2022

galligan commented Dec 11, 2022

lyra-ai commented Dec 11, 2022 • edited

herman925 commented Dec 12, 2022

sluongng commented Dec 12, 2022

sluongng commented Dec 12, 2022 • edited

JoshuaDoes commented Dec 12, 2022

tianlichunhong commented Dec 13, 2022

JoshuaDoes commented Dec 13, 2022 • edited

UnGrosLoup commented Dec 13, 2022

xinmans commented Dec 13, 2022

sluongng commented Dec 13, 2022 • edited by unfurl-links bot

LyghtCode commented Dec 13, 2022 • edited by unfurl-links bot

sluongng commented Dec 14, 2022 • edited by unfurl-links bot

kaixindeken commented Dec 16, 2022

pupubird commented Dec 16, 2022

sluongng commented Dec 16, 2022 • edited by unfurl-links bot

Lordyzagat commented Dec 16, 2022

tianlichunhong commented Dec 17, 2022 • edited by unfurl-links bot

pincombe commented Dec 18, 2022 • edited by unfurl-links bot

TTTq commented Dec 20, 2022 • edited

Use this head to avoid robot detection

permblackshark commented Mar 25, 2023 • edited

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023 • edited

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023 • edited by unfurl-links bot

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023 • edited by unfurl-links bot

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023

acheong08 commented Mar 25, 2023

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023 • edited

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023 • edited by unfurl-links bot

acheong08 commented Mar 25, 2023

GchinoD commented Mar 25, 2023

lyra-ai commented Dec 11, 2022 •

edited

sluongng commented Dec 12, 2022 •

edited

JoshuaDoes commented Dec 13, 2022 •

edited

sluongng commented Dec 13, 2022 •

edited by unfurl-links bot

LyghtCode commented Dec 13, 2022 •

edited by unfurl-links bot

sluongng commented Dec 14, 2022 •

edited by unfurl-links bot

sluongng commented Dec 16, 2022 •

edited by unfurl-links bot

tianlichunhong commented Dec 17, 2022 •

edited by unfurl-links bot

pincombe commented Dec 18, 2022 •

edited by unfurl-links bot

TTTq commented Dec 20, 2022 •

edited

permblackshark commented Mar 25, 2023 •

edited

GchinoD commented Mar 25, 2023 •

edited

GchinoD commented Mar 25, 2023 •

edited by unfurl-links bot

acheong08 commented Mar 25, 2023 •

edited by unfurl-links bot

acheong08 commented Mar 25, 2023 •

edited

GchinoD commented Mar 25, 2023 •

edited by unfurl-links bot