For the complete documentation index, see llms.txt. This page is also available as Markdown.
Advanced Chat
Multi-models, Branches, Long Conversations, and Artifacts
[Conversation] Best for organizing your thoughts while discussing. In addition to single-model Q&A, you can compare multiple models side by side, create branches from any message, manage long conversation context, and keep using files, images, code, and citations from replies as artifacts.
If the task requires continuous local file read/write, calling multiple tools, or running for a long time, switch to the Agent in [Work]. Conversation is better for discussion, comparison, and finalizing, while Agent is better for execution.
Write the question and output requirements in full first; only add model comparisons or message branches when cross-checking is needed.
Choose capabilities based on your goal
Goal
Recommended practice
Compare the views of different models
Select multiple models in the input area and send the same question
Keep the original discussion and explore another approach
Create a branch from a key message
Continue a very long discussion
Check context usage, and summarize if necessary before starting a new topic
Have the next question sent later
Use the message queue without interrupting the current reply
Continue working with files or code in the reply
Open the artifact preview, then download, copy, or continue in a new task
The model selector lets you choose one or more models for the same question.
Recommended order
1
1. First, write the question completely
Explain the goal, materials, constraints, and desired output format. Multiple models will only amplify differences in the original question; they will not automatically fill in missing information.
2
2. Then decide whether comparison is needed
Select multiple models only when you need different perspectives. For everyday Q&A, keep to a single model for a clearer interface and easier usage control.
3
3. Save the useful conclusions
Store reusable materials in notes or the knowledge base; hand over work that needs further execution to the Agent, along with the conclusions already confirmed.
Multi-model comparison and message branching
Compare answers while preserving exploratory paths
