Working with chats

A chat holds one conversation with a coding agent. Each chat belongs to a project and has its own model, permissions, and conversation history. Use separate chats for tasks you want to return to independently, such as fixing a bug and reviewing a feature.

Start a conversation#

1. Select the project tab you want to work in.
2. Click New chat in the left toolbar.
3. Check the model and permissions below the message box.
4. Describe the task and submit your message.

New chats use your default model settings. An empty draft may be reused when you click New chat again before sending a message. Once you send a message, the conversation appears in the project's history.

For a useful first message, give the agent an outcome and a way to check it. For example:

Find why the settings form loses its values after a refresh. Fix the cause and explain how you verified the fix.

Use file references and attachments to point to relevant code or an image.

Return to a conversation#

1. Click Chat history in the left toolbar.
2. Browse the current project's chats, with recently updated conversations first.
3. Enter part of a chat's title in Search history to narrow the list.
4. Select the conversation to open it.

History search matches titles, so a descriptive name makes work easier to find. It does not search the text of every message. If a chat is missing, check the selected project and then Archived Chats.

When multi-chat is enabled, selecting a conversation can bring it into the side-by-side layout. With multi-chat disabled, you view one conversation at a time.

Rename a chat#

Dream can generate a title after a conversation starts. To give it a more specific name, open the chat's … menu and choose Edit, or double-click the title. In Edit chat, enter the name and click Save. Use Cancel to keep the existing name.

Names such as “Fix settings persistence” and “Review settings tests” are easier to distinguish than several chats named after the same project.

Follow work in progress#

While an agent responds, the chat shows its progress and tool activity. Open tool details to inspect commands, file reads, or changes. When the agent supplies a task list, the task control beside the message box shows its progress.

History and project tabs include status indicators for work that is processing or has finished; project tabs can also indicate that an agent is waiting for your answer. Select the project and open the conversation to read its latest message. Use the stop control in the composer to interrupt a response. Stopping a response does not undo file changes already made.

After a reply finishes, use View file changes from this reply in its footer to see the files it changed and restore any of them.

Close, archive, or delete?#

Action	Result	How to return
Close a chat column	Removes it from the multi-chat layout	Open it from Chat history
Archive a chat	Removes it from active chat history	Restore it in Settings → Archived Chats
Delete an archived chat	Permanently removes the chat and its messages from Dream	It cannot be restored through Dream

Closing a column is useful when you need more space. Archiving is useful when a task is finished. Neither action reverts the agent's changes to project files.

To try a different approach while keeping the conversation so far, branch from a message. To prepare a task for later, save a prompt in Stash. For instructions you want to reuse across projects, create a reusable prompt and run it from the composer's + → Run prompt menu. Type $ to invoke an agent skill.