Chat with AI
Chat with AI lets you build and modify your website through a natural language conversation with an AI assistant. Send instructions to replace content, adjust styles, add sections, or diagnose issues — the AI processes each task step by step and shows the entire generation process in the conversation.
Make sure that your workstation connection is active before you send messages. If the connection is unstable, messages may be lost without entering the conversation.
Send messages
Click the input bar at the bottom of the conversation area and enter your request.
Send the message in one of the following ways:
Press Enter.
Click the send button on the right side of the input bar.
Click the small arrow next to the send button to switch the keyboard shortcut for sending messages.
View the AI response in the message stream. For a complex task, the AI first breaks the task down and then runs it step by step. You can watch the processing in the message stream.
Describe your request clearly
Cover the following elements in your request so that the AI understands it accurately:
| Element | Description |
| Target | The page or module to modify, such as the homepage, a product details page, or an article list. |
| Specific request | The result that you want, such as replacing copy, replacing an image, or adding a section. |
| Style requirements | For visual changes, the style, color scheme, and other preferences that you want. |
Example:
Replace the "Company profile" section at the bottom of the homepage with a "Partners" section that shows partner logos, and keep the overall style clean.
You do not need to describe everything at once. The AI supports incremental generation. Start with the overall request, and then refine it step by step in later messages.
Generation and response
After you send a request, the AI analyzes it and starts generating. The entire process is visible in the conversation.
Control generation
Stop generation: Click the stop button at any time during generation to interrupt the current task.
Confirmation card: For critical operations, the AI shows a confirmation card in the conversation and continues only after you confirm.
Self-check report: After generation finishes, the AI automatically runs checks such as compilation and page accessibility, and reports the results in the conversation.
Read the response
Message footer: Shows the mode, model, "AI website building", and elapsed time.
Processing details: Next to the message that you sent, click Processing details to expand and view how the AI reasoned about the task.
Reasoning, tools, and file changes: Messages for complex tasks show markers such as "Thought for Xs · N tools · N files".
Conversation modes and models
Conversation modes
The conversation mode determines whether the AI modifies files and whether it asks for confirmation midway. The input bar uses Execute mode by default.
To switch the mode
Click the mode button in the lower-left corner of the input bar. The button shows Execute by default.
Select the mode that you want from the drop-down list. Each mode includes a description of its use case.
Available modes
| Mode | Description | Modifies website files | Scenarios |
| Execute | Runs tasks, and can modify files and run tools. | Yes | Daily operations tasks. |
| Managed | Delivers the task end to end without repeated confirmation. | Yes | Batch tasks and scenarios where you want the AI to work continuously without interruption. |
| Analyze | Only analyzes, and does not modify files directly. | No | Diagnosing problems, reviewing content, and cases where you only need suggestions. |
Managed mode runs directly without midway confirmation. Define the task boundaries clearly in your instructions so that the AI does not do more than you expect.
Select a model
You can switch models based on the task.
Click the model name button in the input bar to expand the model list.
(Optional) Enter a keyword in the Search models box at the top of the list to filter models. For example, if you enter
qwen, the list shows only matching models.Click the model that you want. The model name in the input bar updates immediately.
Connect a provider
You can also enter an API key to connect models from an external provider.
Open the model list and click Connect provider on the right side of the Search models box.
Select the provider that you want to connect. The following providers are supported:
Provider Description DeepSeek direct connection Connects directly to the DeepSeek API. Alibaba Alibaba Cloud Model Studio (China site). On the provider connection page, enter the corresponding API key as prompted and click Save connection.
Swarm mode
The AI can split a large task into multiple subtasks and process them in parallel.
Trigger swarm mode
State the parallel request in natural language in the conversation. The console does not provide a dedicated swarm mode switch. For example:
"Use swarm or parallel mode to analyze the section structure of the homepage and the product list page at the same time. Analyze only, do not modify."
Watch the execution
During parallel execution, you can track progress in the conversation area:
Subtask cards: For a large task, subtask cards with individual timers appear in the conversation area. Each card shows the subtask name and elapsed time, so you can watch the progress of each subtask in real time.
Parallel file operations: The AI modifies multiple files in batches, and the message shows the number of file changes.
Concurrent tool calls: The AI may call multiple tools within a single message.
For a small task, the AI may complete the work directly without splitting it into subtask cards. Whether subtask cards appear depends on the size of the task.
Upload files
You can attach local files to the conversation for the AI to read and reference.
Click Attach file (+) in the input bar.
Select local files. You can select multiple files of different types at a time.
After the files are attached, file cards appear above the input bar. Send them together with your message. Text, image, Office document, and other formats are supported.
Manage attachments: Before you send the message, hover over a card to remove the attachment, click a text card to preview the full text, or click an image card to view a larger version.
Where files go: The AI can read text files directly. Images are automatically saved to the media library and marked as added to image management. Office documents are stored locally on the workstation.
Recognition depends on the model: Some models can read documents but cannot read text in images. Models labeled Image recognition supported can recognize text in images. Switch to a model that supports image recognition before you ask the AI to read an image.
Image recognition tips: Image recognition requires a high-resolution image with clear text. If the text is too small, the image resolution is considered too low. You can also send only attachments without any text, and the AI interprets the attachment content on its own.
Images that you attach in a conversation are automatically added to the media library, and you can select them directly when you later replace images on your website. For more information about the media library, see File management.
Manage conversations
Start New conversation, Share conversation, and Compact context from the conversation options menu by clicking the … button in the lower-left corner of the input bar. Start a rollback or a restore from the buttons next to each AI message.
Start a new conversation
Starting a new conversation clears the current chat history and opens a fresh conversation.
Starting a new conversation clears the current chat history, and the historical messages cannot be recovered. You also cannot retrieve the earlier conversation by rolling back. Confirm before you proceed.
The website content is not affected. Only the conversation history is cleared.
Click the … (conversation options) button in the lower-left corner of the input bar, and then click New conversation.
In the confirmation dialog box, read the message: "The chat history will be cleared. The website content is not affected. Historical messages cannot be recovered."
After you confirm, click OK.
A Rename conversation button is also available next to the conversation title. Use it to change the name of the current conversation.
Compact the context
When a conversation becomes too long, the AI responds slowly, or the response quality drops, you can compact the context. The AI condenses the earlier conversation into a structured summary and frees up context capacity.
View context usage
Hover over the context usage icon, a ring-shaped progress icon, in the input bar to view the current usage percentage.
Click the … button in the lower-left corner of the input bar, and then click Compact context.
Compaction requires no confirmation. It starts immediately after you click.
Compaction completes as an AI message. A message labeled compaction is appended to the end of the message stream. Its content is a structured summary that contains Objective, Important Details, and Work State, which cover the core information of the current conversation.
Compaction results
Context usage drops significantly, for example from 6% to 2%.
After compaction, scroll up to view the earlier messages, which remain fully visible.
Key project states, such as website settings and completed items, are retained in the summary, so the AI can continue to work based on this information.
Compacting the context permanently changes the context state of the session. It runs immediately after you click, with no confirmation step. However, compaction does not change the website content, and historical messages remain viewable.
Roll back
If a change made by the AI does not meet your expectations, you can roll the conversation and the website back to the state before a specific message. If the rollback turns out to be wrong, you can restore the previous state.
A rollback removes conversation messages and also rolls back the website files.
Roll back to here
Each AI response has a Roll back to here button on its right side. The button is highlighted when you hover over the message.
Find the AI message that you want in the message stream and hover over it.
Click the Roll back to here button on the right side of the message.
The rollback runs immediately. The message and everything after it are removed, and the website files are rolled back to the state before that message.
Roll back to here has no confirmation dialog box. The rollback runs as soon as you click. The rollback discards conversation content and rolls back website files at the same time. Confirm the target message before you click.
Undo a rollback
After a rollback, a banner appears at the bottom of the conversation area: "Rolled back to here. Sending a new message discards the content after this point." The banner includes a Restore button.
Click Restore to undo the rollback. The conversation and the website pages return to the state before the rollback, and the banner disappears.
You can roll back multiple times in a row. There is no limit on the number of rollbacks.
The Restore button works only if you have not sent a new message after the rollback. Once you send a new message after a rollback, the content removed by the rollback is discarded permanently and cannot be retrieved, and the Restore button disappears. After a rollback, decide whether you need to restore before you send a new message.
FAQ
Nothing happens after I send a message. What do I do?
If the workstation connection is abnormal, the message that you send may be lost: the input box is cleared as usual, but the message does not enter the conversation. If you get no response after you send a message, refresh the workbench page and try again.
Why does the AI say it cannot find the file that I uploaded?
In rare cases, the AI incorrectly reports that the file does not exist. Explicitly ask the AI to read the attachment and send the message again, for example: "Read the attachment that I uploaded and answer the question."
Why can the AI not read the text in my image?
Check two things. First, switch to a model labeled Image recognition supported. Second, make sure the image is a high-resolution image with clear text. If the text is too small, the image resolution is considered too low.
What is the difference between Execute mode and Managed mode?
Both modes modify files. Managed mode does not ask for confirmation midway, which suits batch tasks. Define the task boundaries clearly in your instructions.
Does the website content get lost after I start a new conversation?
No. Starting a new conversation clears only the chat history. The website content is not affected.
Does compacting the context lose historical messages?
No. You can still view historical messages after compaction, but compaction permanently changes the context state and cannot be undone.
Can I restore the previous state after a rollback?
Yes, but once you send a new message after the rollback, the rolled-back content is discarded permanently.
What can others see through the link that I share?
The full conversation content. The link is fully public. Exercise caution when sharing conversations that contain sensitive information.
What is the difference between Roll back to here and New conversation?
Roll back to here rolls the website files back to the state before a specific message. New conversation clears only the chat history and does not affect the website content.