The video on this page shows the Chinese interface.
Start and continue
On desktop, go to Projects to create a task. Choose the engine, model and device, and send it with your first request. For a one-off task you can leave No project selected. To work on existing material, go to the matching project and start from a specific folder.Turn these experiment results into a table and mark missing items. Don’t modify the raw data yet. Save the result separately as Markdown.To add to an existing task, go back to it in the sidebar and send more requests. A new task starts working as soon as you send the first message. A forked task inherits the context and continues only after you give it a new request. Among the modes for a new task, Work communicates through chat and process summaries, and Code shows the Agent’s terminal interface. Both can do the work. Which modes are available depends on the current engine and device.
Watch the process, open the results
Reasoning and tool use are collapsed into a one-line summary. Click it to see each step. File links open at the relevant location. Supported files can be previewed beside the chat. While the Agent is working, you can add your next message. Queued messages appear near the input box. Edit, delete or reorder them with the actions the interface offers.Revise requests, fork, and side tasks
Message editing is available only when the current Agent and its history support it. It may be unavailable while the Agent is working or messages are queued. Save any result you need to keep first. Forking the context does not copy a running terminal.