Editing with AI

Open AI Chat from the left rail, or pop it into its own window with Open chat in its own window at the top of the panel. You need to be signed in.

AI Chat — ask for edits, attach screenshots, review the stream.

Not sure where something is in Plyed? Ask the chat. It can explain and point to the control; see Ask Plyed How To….

Ask for changes

Type what you want and press Enter. The agent reads and edits your project’s files directly. You’ll see its progress (thinking, file reads, edits) stream in, and the canvas refreshes as it works.

Enter sends. Shift+Enter adds a new line.
The stop button (Stop response) ends the current reply.

Steer a reply in progress

You don’t have to wait for the agent to finish. Type a follow-up while it’s working and press Enter. The message goes straight into the running reply and the agent picks it up at its next step. Use this to redirect it, add a detail or correct course.

To hold a message until the current reply is done instead, press ⌘Enter. It waits in a queue above the composer and is sent when the reply finishes.

A few messages always wait in the queue rather than steering: messages with images attached, and messages in a cloud agent chat.

Attach a screenshot

Drop an image onto the chat composer, or paste one with ⌘V (handy for screenshots). Thumbnails appear above the text box. Remove any with ✕.

You can attach up to six images per message, 4 MB each, in PNG, JPEG, WebP, GIF or TIFF format. Pick a vision-capable model in the model dropdown when the image matters, because not every model can see images.

Attach a folder

Click Attach folder in a project chat to choose a folder on your computer, or drop a folder onto the message box. If Plyed can’t identify a dropped folder, it opens the folder picker.

For example, attach a downloaded 3D model folder and ask:

Inspect this model and its textures, prepare it for WebGL, and use it in our scene.

The agent can read the folder and copy the files it needs into your project before working on them. Your original folder stays unchanged. Whether a conversion works depends on the file format and the tools available to the agent.

Attached folders stay with that chat when you reopen it. Click × on a folder chip to detach it. Files already copied into the project stay there. Folder attachments only work in chats on your Mac, so remove them before switching a chat to a cloud agent. In Plan mode, the agent can look at attachments but can’t import or change them.

Work in several terminals

The agent can run several commands at once, for example a model conversion alongside a texture check. Ask it to run independent work in separate terminals. It can read their output, type input into them, and close each one when it’s done.

In a project chat, Agent terminals shows the running commands and their status. Select one to read its output, or stop that command on its own. Stopping one doesn’t stop the others or your site preview.

These terminals only last for the current reply. Plyed closes any unfinished command when the reply ends, when you stop the reply, or when you quit the app. Very long output is trimmed to its most recent part. In Plan mode, the agent can read terminal output but can’t start commands or type into them.

Let it check its work in a browser

The agent can open a real browser next to the chat, click through your site the way a visitor would, and take screenshots to confirm a change actually works. It does this on its own when it decides a visual check is worth it. To open the browser yourself, click Open browser (the arrow in the top corner of the chat).

The agent only sees those screenshots on a vision-capable model. On a model that can’t see images, the agent is told it didn’t see the page, so it doesn’t guess what’s there.

Background tasks

When a job splits into independent pieces (say, two unrelated edits or two things to look up), the agent can hand pieces off to background tasks that run alongside the main reply. Running tasks are pinned above the composer so you can always see them. Up to four can run at once.

Point at an element

The fastest way to talk about something specific: click it on the canvas, then click Ask AI in the Inspector. The chat opens with that element attached as a small chip showing its tag and where it lives, and the agent works on exactly that spot. Remove the chip with ✕ if you change your mind.

Project docs (styleguides, plans, rules)

Give the AI lasting context, such as your brand voice, design notes and project plans, from the Docs panel in the left rail:

Kind	Where it lives	What the AI does with it
Rules	`.plyed/rules/*.md`	Always included with every message (keep these short).
Docs	`.plyed/docs/*.md`	Listed for the AI to read when relevant (styleguides, `design.md`, plans).

Drop a .md file onto the panel, or onto the window. Markdown files go to Docs, not Assets.
Click a file to read and edit it in the panel (⌘S saves), or click Code to open it in the code editor.
Attach (or @ on a row) adds that doc to your next chat message so the AI reads it before editing.

To change how a single file is used, add apply: always or apply: optional to its frontmatter.

For playbooks you want in every project, or only for one message, use skills: type $ in the composer to attach one.

Modes and models

Two dropdowns sit under the composer:

Mode
- Default can change anything.
- Plan is read-only: ask questions and get a plan, and nothing changes.
- Web focuses on your site’s pages and components.
- CMS focuses on your content.
- Stylesheet focuses on your styles and design tokens (colors, type, spacing).
Model: pick a model, or leave it on auto. Bigger models are smarter and slower. All of them use the same AI credits, and premium models use them faster. On the Free plan the AI always uses the default model. See Plans & Credits.

Several chats at once

You can have up to 10 chats open. Start a new one with New Chat (⌘N). Each chat runs on its own, so you can start a long task in one and keep working in another. Chats with a reply in progress are marked, and a chat catches up when you switch back to it. Chats are kept when you quit and reopen Plyed.

In the chat sidebar, right-click a chat to rename, pin, archive or delete it. A chat that’s still replying can’t be deleted.

Long chats keep working too. When a conversation grows too long to send in one go, Plyed automatically summarizes the older part so the agent keeps the context without you doing anything.

Review and undo

Each reply shows the edits it made as diffs. If a reply didn’t go the way you wanted, revert these edits puts every file it touched back.

Rewind to a checkpoint

Plyed also saves a checkpoint of your whole project before every reply that changes files, in project chats (account-scope chats have no project to checkpoint). Open Checkpoints in the chat sidebar (below New Chat) to see them, newest first. Each one shows the message that followed it and how long ago it was taken.

Click a checkpoint to put every project file back the way it was at that point. You’re asked to confirm first. Plyed saves your current state as a “Before restore” checkpoint just before restoring, so the notice that appears afterwards lets you Undo. Checkpoints are stored outside your project folder, so they never touch your own git history.

Checkpoints use Apple’s git tool. The first time on a new Mac, you may see a system prompt to install the command line developer tools — see Install & Sign In.

Checkpoints is unavailable while a reply is running in that project.

Self-checking

After a reply that changed files, Plyed watches your site preview and its browser console for a moment. If new errors show up, a banner above the composer says New errors appeared after the last edit. Choose Ask AI to fix to send those errors to the agent, or Dismiss to ignore them. Plyed never sends them on its own: the errors come from your site’s code, so you decide whether the agent acts on them. Sending any other message also clears the banner.

If your project has Astro’s type and template check (astro check) installed, Plyed also runs it after a reply that edited files, and lets the agent fix any errors it reports once. You don’t have to wait for it. The composer is free as soon as the reply is done, and Checking your changes… shows under the reply while the check runs. If the agent fixes something, the fix arrives as its own diff a moment later. The check never blocks or undoes your edits, even if it can’t finish.