Chatting with the agent

You build your app by chatting with the agent in Newly v2, at app.newly.app and in Newly for Mac. The prompt box works the same way in the web app and in the Mac app.

​

Send a message

1. Type what you want in the prompt box.
2. Press Enter to send. Press Shift+Enter to start a new line.

Write in any language. The agent replies in the language you write in, and keeps the code itself in English unless you ask otherwise. For how to phrase requests, see Prompting tips. To attach screenshots or files, see Screenshots, files and voice.

​

Which AI builds your app

• Remote chats always use Newly Agent. Remote chats run in the cloud, and every chat in the web app is a Remote chat. Newly chooses the model behind Newly Agent and may change it. There’s no model or effort picker for Newly Agent.
• Local chats in the Mac app can also use your own Claude or Codex login. See Local chats on your Mac.

​

While the agent works

The send button turns into a stop button, and a status line shows what the agent is doing and how long it has been working, for example “Building the app”. After about 10 seconds, the status line adds “You can stop safely: completed work stays saved.” To see the code the agent wrote, expand a step in the chat. Steps that create or change a file show the code or the diff, with the file path at the top.

​

Stop a turn

Press the stop button to end the current turn. The chat itself stays open, and your next message continues from where the agent stopped. The stopped reply is marked “Stopped - completed work is saved”. What happens when you close the tab or the app depends on where the chat runs:

• Remote chat: closing the browser tab or the Mac app doesn’t stop the turn. The agent keeps working in the cloud. When you come back, the chat shows the turn still running or finished.
• Local chat (Mac): the turn ends when the Mac app closes or your Mac stops. The reply is marked “Interrupted - your work is saved”.

​

Queue more messages

You don’t have to wait for the agent to finish. While a turn runs, the prompt box reads “Add to queue…”.

• Messages you send now wait in a queue. A chip above the prompt box shows how many, for example “2 messages in queue”.
• Queued messages run one at a time, in order. Each one runs in the mode you picked when you sent it. See Build, Plan, Ask and Design.
• Open the chip to see the queue. You can remove any message that hasn’t started.
• Stopping a turn also pauses the queue, and the chip shows the queue as paused. Choose Resume to continue with the queued messages, or Discard to drop them. If you send a new message instead, it runs first and the queue continues after it.

​

Answer the agent’s questions

Sometimes the agent needs a decision before it can go on, for example whether other people should be able to sign in and see the same data. When that happens, the prompt box turns into an answer panel:

• A label such as “1 of 3” shows which question you’re on. Use the back arrow to return to an earlier one.
• Pick one of the offered answers, or write your own in “Type your answer…”.
• Press Skip if you don’t want to answer that question.

The normal prompt box comes back after you answer or skip. Time the agent spends waiting for your answer doesn’t count toward the one-hour turn limit.

​

Copy, delete and other chat actions

• Copy a message: hover over it and use the copy button.
• Copy chat history: in the chat’s options menu.
• Delete chat: in the chat’s options menu. See Delete a chat, project or account.

You can’t edit or retry a message after you send it. To take back changes, use Restore to here. See Undo and Restore to here.

​

View-only access

If someone shared a project with you as Can view, you won’t see a prompt box. Instead you see “View only” with the message “You can read this project and watch what the agent does. Ask its owner for edit access to send prompts.” To send messages, ask the owner for Can edit access. See Share a project.

​

Credits

Newly Agent work uses credits, based on how much work the agent does in a turn. This includes turns the agent starts by itself to fix a build or runtime error. See How credits work.

​

Related