1. Go to this link and sign in or sign up for a Cloudflare account.
2. In the sidebar, select Workers & Pages.
3. On the page that opens, click Create application.
4. Choose Create Worker.
5. Give your worker a name and click Deploy.
6. On the worker detail page, click Quick edit.
7. Copy all the code from here and paste it over the existing code in worker.js. Then click Save and deploy.
8. Copy the worker domain that looks like xxxx-xxxx-xxxx.xxxx.workers.dev (not a URL like https://xxxx-xxxx-xxxx.xxxx.workers.dev/, please remove the prefixes and suffixes) and paste it as Wss Domain in the settings dialog of SydneyQt. Then click Save.

• Wss Domain: Used to proxy websocket interface, break regional restrictions.
• Proxy: The proxy used to access New Bing, recommended to be an http proxy, such as Clash's 7890 port. If you use a Wss domain that is reverse-proxied by Cloudflare, you may not need a proxy to connect, but since the HTTP GET endpoint for creating conversations is still blocked, you still need a proxy.
• Dark Mode: Imported a custom css from Python Qt to implement dark mode effect, some minor rendering problems may occur on some UI, such as text overflowing buttons, etc.
• Conversation Style: New Bing provides three chat modes, namely Creative, Balanced, and Precise. Among them, Creative and Precise modes are backed by GPT-4, and Balanced mode is backed by GPT-3.5. It is recommended to use Creative mode.
• No Suggestion: New Bing will generate three suggested user responses based on AI's output results. After checking this, the suggestion bar will not be displayed, but AI will still generate suggestions, which means that you have to wait for a while after each round of message sending ends. This can't be turned off even by modifying optionsSets.
• No Search Result: There are currently two ways to disable search: instructing in the jailbreak prompt and automatically adding the "#no_search" keyword after each user-sent message. This option uses the latter.
• Font Family and Size: Font family and size settings for context box and input box.
• Stretch Factor: Used to adjust the placeholder ratio of Chat Context and User Input input box, which is an integer. The larger this value, the higher the Chat Context, and correspondingly, the smaller the User Input height.
• Suggestion on Message Revoke: Due to Microsoft's restrictions, AI may suddenly realize that something is wrong after outputting a piece of content, and then withdraw the message and apologize. Of course, revoking in a third-party client is invalid, at most it means that subsequent content cannot be output. But at the same time it will not generate reply suggestions either. Therefore, the text here is used to replace the suggestion bar display text at this time. The default is Continue from where you stopped, instructing AI to continue outputting. Since the new message sent will attach the chat record context in webpage_context, it will not go through external censorship, so AI can continue writing on the content that was just interrupted, unless there is sensitive output again in the continued content.
• Revoke Auto Reply Count: If the value is not 0, then when a message withdrawal is detected, it will automatically send the "message revoke suggestion" text to let AI continue writing. The maximum number of times sent will not exceed the value set here.
• Send Quick Responses Straightforward: There is a Quick button on top of the input box for quickly sending some template texts. Such as "Translate the above text into Chinese" and so on. When this option is activated, if you click on one of the template texts in Quick and there is no text in the input box, it will directly send the template text to AI; and if there is text in the input box, it will add the template text below the existing text.

Below are some ChatGPT related settings, because SydneyQt supports OpenAI's API:

• OpenAI Key: API key, usually starting with sk-, but the program will not validate it.
• OpenAI Endpoint: Custom OpenAI API endpoint, useful when using third-party distributors, such as openai-sb.com in China provides API that is much cheaper than official ones. It needs to end with /v1.
• Short Model & Long Model & Model Switching Threshold: Now GPT-3.5 supports 4k and 16k two models, and the two models charge differently. How to minimize costs as much as possible? Of course it is long text with long model and short text with short model. Model Switching Threshold is a token count. If the current Chat Context's token count is greater than this value, then use Long Model for the next request sent; otherwise use Short Model.
• Model Temperature: The model's temperature, between 0 and 2. The larger the value, the more random the model's output. Usually keep default.