Agent mode is a more careful way for Phluffle to handle your prompts. Instead of one pass, it plans the work, builds it step by step, checks the result and fixes what it finds. Use it when you want fewer broken links and missed screens, and you don't mind waiting a little longer.
Pro Agent mode is part of Pro. New accounts get a 48-hour free trial.
Choose between Chat and Agent
You pick how Phluffle works in Settings → Preferences, under How Phluffle works with you:
| Mode | What it does |
|---|---|
| Chat | Quick back-and-forth. Each prompt is one pass: you see what changed and keep going. |
| Agent | Plans the work, builds it step by step, then checks its own work and fixes what it finds. You watch every step with timings. Takes a little longer and uses a few more credits. |
You can switch any time. The change applies to your next prompt.
In the editor, a lime Agent pill next to the prompt box shows when Agent mode is on (it says · trial during your free trial). Click it to jump to Settings.
Your free trial
New accounts start in Agent mode with a 48-hour free trial. While it runs, Settings shows Free trial · {time} left and the note "After that you're back to Chat, unless you go Pro."
When the trial ends, Settings says "Your free trial of Agent mode has ended." and your prompts go back to Chat. Click Go Pro to use Agent to keep using it. See Plans and credits.
Good to know Members of a Pro Enterprise workspace are Pro too, so they can use Agent mode. See Team workspaces.
What happens when the agent works
Send a prompt as usual. A live card appears in the chat, starting with Planning the work…, and lists each step with its timing as it goes:
- Plan the work. The agent breaks your request into steps.
- Build. Usually one step per screen. For a short request it may be a single step, such as Build the prototype (for a new prototype) or Make the requested changes (for a change).
- Check links, screens and accessibility. The agent looks over what it built.
- Fix N issues. Only if the check found problems it can fix.
- Generate N images. Only if the prototype has image placeholders and AI images are available. See AI images and the Image studio.
Screens appear in the preview as they're built, just like in Chat mode.
You can leave this page; the agent keeps working. To cancel, click Stop. As with any stopped generation, you only pay for the part that was written. See Versions, restore and retry.
Read the report
When the agent finishes, the card becomes a short report on the new version:
- Done in m:ss: the total time.
- Each step with how long it took.
- Self-check: X of Y passed, plus · fixed N issues if it fixed anything.
The result is a normal new version. You can open older versions, restore them, hand-edit with Visual Edit, or keep prompting.
What Agent mode costs
Agent mode uses credits like any prompt, based on what the AI actually used. Because it plans, checks and sometimes fixes, it uses a few more credits than the same prompt in Chat. Failed generations are free.
When to use it
- Agent is a good fit for new prototypes, multi-screen changes, and anything where broken links or missing screens would slow you down.
- Chat is faster for small, quick tweaks. For changes to one screen or element, also try Scope. See Change one screen or element.