The Prompteryx app

When you sign in, you land on the app's home. Type what you want in the message box and send it. You don't need to choose a tool first: the app works out whether you want an answer, a website, an image, a document, a browser task or one of your connected apps, and picks a suitable model (see Models and Auto).

The buttons under the message box

On the home, a row of buttons sets the kind of task: Do a task, Build workflow, AI Browser Agent, Research a topic, Design, Motion design, Build a website, Make a game, Create an image, Create a video, Create a voiceover, Create music, Create sound effects and Deep research report. Choosing one changes the heading, the hint in the message box, the examples and the model picker. Press the × on its chip to go back to Auto.

Without a button, the app decides. A short line under each reply says what it chose, for example Routed to Website · Claude Opus 5.5 · change; press change to try it another way or turn the result into something else. Documents, slides, spreadsheets and animations don't need a button: just ask. See What you can make.

The message box

• Attach files: images, PDFs, documents and spreadsheets. You can also paste an image. The chat reads them as part of your request.
• AI Browser Agent (the globe): turn the agent on for this message. See AI Browser Agent in the chat.
• Connectors: browse and connect 3,000+ apps (on the home, the strip of app logos). Once connected, just mention the app in your message. In a chat, the plug opens Connectors in this chat, with a switch to turn each app off for that chat. See Connectors.
• Skills (the lightning icon): pick one of your skills, or type /name, such as /meeting-notes.
• Incognito (the eye icon): the chat isn't saved and memory is off.
• Memory off (the brain icon): keep the chat, but nothing is remembered from it or used in it.
• Schedule: run the message on a schedule instead of now. Send becomes Create schedule. See Schedules.
• Model picker: Auto, or a model you choose. See Models and Auto.
• Voice (the microphone): speak instead of typing.

A reply that is still running carries on if you open another chat or close the tab, and is there when you come back. Press Stop to end it early. Each chat keeps its own unsent draft, attachments included, so you can leave a half-written message and come back to it.

Inside a chat

• Previews: sites, pages, games, documents and slides open in a preview next to the chat, with Download, Publish and Share to Explore.
• Progress: the button at the top of the chat lists everything it made (Outputs) and what it used (uploads, connected apps, browser, memory, skills and models).
• Share: send someone a read-only copy of the chat with a share link.
• Sub-agents (early access): for a big job, the chat can hand parts to helpers that work at the same time. While the reply runs, each helper has its own row with its task, what it is doing, its time, credits and model; the rows stay with the reply, folded, and Progress has a Sub-agents row with the total.

If a page ever breaks, you see Something went wrong with Reload, Try again and Go to home. Reload fixes most cases, including right after we release an update. A short, anonymous error report is sent to us so we can fix it: it never includes your e-mail address or what you typed.