visual feedback for AI-built apps

Client sign-off for AI-built apps

Your client comments on the live preview, no login. You get a clean punch-list your AI agent can fix.

<script src="https://room.patchrooms.com/v1/patchrooms.js"
        data-project-key="pr_xxx"
        data-mode="artifact-review"></script>

Local mode runs in your browser: no account, no key, nothing leaves the page.

AI made coding faster.
Feedback got messier.

Screenshots in one chat, bug reports in another, a plan buried in a thread, the diff somewhere in a PR. By the time context reaches the person (or the agent) fixing it, half of it is gone.

Every issue gets a room

One click opens a room around the problem and keeps everything that matters in one place.

A room is one report: a single piece of feedback and its thread. Rooms live in a project, one per app or repo.

Screenshots & annotations

Point at the exact element. Pin a comment to it, capture a screenshot, or quote on-screen text so nothing gets lost in translation.

Voice notes, auto-transcribed

Too much to type? Hold to record. Patchrooms transcribes your voice into clean, agent-ready text. No typing, no copy-paste.

Comments & statuses

Humans and agents share one thread per report, with a status from new to fixed, so nobody digs through chats to find out what happened.

Agent-ready export

Every room exports as clean Markdown your agent can act on: copy it as a prompt, download it, or let your agent pull it over MCP.

Send better context to your agent

A room is the context an agent actually needs. Hand it to Claude Code or Cursor as clean Markdown so it reproduces, patches, and ships with taste instead of guessing. Or let agents read reports directly over the MCP server, available now.

More on agent-ready feedback.

Built for tiny teams that move fast

No corporate workflow engine. Just rooms where humans and agents share the same context, so a handful of people can review, patch, and ship without losing the thread.

What your agent receives

Not a ticket. Patch context.

Every report exports as clean Markdown: selector, page, viewport, console state and the ask. Or your agent reads it straight over MCP. No interpreting screenshots.

This is one real export.

## Feedback: Checkout submit button

Element: button[data-patchroom-id="checkout-submit"]
Page: /checkout · 1440×900 · Chrome 126

Issue
Stays clickable while required fields are empty and
submits an invalid form.

Expected
Disabled until the form is valid: reduced opacity
+ aria-disabled="true".

Console
POST /api/order 422 (validation_failed)

Built for every AI coding agent

Your agent ships it: Claude Code, Cursor, Lovable, and more. Patchrooms closes the feedback loop.

Drop one script with data-mode="artifact-review" on any AI-generated preview and your team comments right on the pixel. Patchrooms turns it into clean, agent-ready context your AI tool can fix.

One feedback layer. Many ways to run it.

From hands-on review to a timed loop where your agent fixes and closes reports on its own. Pick the autonomy that fits the change.

  • Review agent proposes, you approve
  • Testing / QA reports as a checklist
  • Auto-fix loop timed, hands-off
  • Spec-writing vague feedback → a plan

See it in action

Drop the SDK on your app, gather feedback in rooms, and send sharper context to your agents.