01 / 06
One line on your site.
Copy the script tag, allow the domain, done. No npm package, no build step.
Script tagAny platformTwo minutes
BEFORE
Slack14WhatsApp9Telegram6Phone call3Email27reads everything, types it all upTickets by hand · no context
?
Five channels, one person, tickets by hand.
Your users click on the problem. The AI spots the pattern.
Feedbapp collects feedback directly in your application, whether website, web app or software: with a screenshot, the clicked element and the technical details. From this context, the AI classifies every report and groups matching topics, in about five seconds.
Closed beta. We open it up in small groups, you will hear from us within a few days. No spam.
WITH FEEDBAPP
shop.yourcompany.com/order/4711
Order #4711
DeliveryWed, Sep 16
StatusDelivered
What's wrong here?
Says “delivered”, but the package never arrived.Screenshot · Element · ConsoleSend
about 5 seconds
BugOrder status jumps to “delivered”
Topic “Order status” · 3 similar reports
Straight from the product. Sorted automatically.
The whole thing in one minute
From the first click to the plan.
Four people report, the AI sorts, you approve, your coding agent plans and gets going.
Other tools get a sentence. Feedbapp gets the whole scene.
An AI can only classify as well as what it is given. Tickets and forms deliver a sentence, and after 200 entries you have a list, not a picture. Feedbapp sees the screenshot, the clicked element, the console and the failed requests.
What others see
“The export doesn't work”
What Feedbapp sees
- Element
- [data-testid="export-button"]
- · Page
- /reports
- Console
- TypeError: t.rows is undefined
- Network
- 500 · POST /api/export
- · Window
- 1440 × 900
Type Error · Topic “CSV export fails (3)”
How it works
From click to topic. In six steps.
02 / 06
The user says it right where it happens.
Button bottom right, four categories, one sentence. No account, no extension.
No account neededBug · Idea · Feature · ProcessOn every page
03 / 06
It points at the problem.
Screenshot, the exact element clicked, console messages and failed requests go along automatically. The widget redacts known tokens and keys first.
ScreenshotElement markerConsole & network
04 / 06
The AI turns it into a topic.
Type, summary and assignment to an existing topic, in about five seconds, without anyone tagging it.
ClassificationTopic clusters~5 seconds
05 / 06
You decide what happens next.
Six columns from “Future” to “Done”, topics sorted by weight: you drag the card, the AI only pre-sorted it.
Board & listTop topicsNotes in the history
06 / 06
Your agent does the rest.
Claude Code or Cursor fetch the open topics, work through them and reset the status, in a loop, without you sitting in between.
MCP serverSix toolsAutopilot loop
$ npx -y -p feedbapp-cli@latest feedbapp-mcp setupJoin the waitlist
Closed beta. Reply within a few days, no newsletter.
Also built for QA, your own team and staging.
QA
Testers report straight from the interface: screenshot, element and page are already attached, no ticket typed by hand. The AI sorts it by type and topic.
Dogfooding
Your own team uses the product and reports from inside the application. That is how Feedbapp itself runs.
Staging and internal
The widget can be embedded on a staging or internal address. The origin is allowed per project, in the project settings. With no entry, every origin is allowed.
We build Feedbapp with Feedbapp.
The orange button bottom right is our real widget, the same one our customers embed. Whatever you report there lands classified in our board. If it is good, it goes into the next plan.
Your coding agent can read the board and write back to it.
One command, then Claude Code knows your topics.
$ npx -y -p feedbapp-cli@latest feedbapp-mcp setup“Show me the open P0 bugs from this week.”“Summarize the CSV export topic.”“Set the ticket to in progress, with a note.”
In the loop: Because the agent can also set the status, the loop runs by itself: it picks the next topic, builds it, marks it done and moves on. The agent keeps building while you work on something else. You review instead of coordinating.
Six tools for Claude Code, Cursor and Codex, on every plan, including Free. → Docs
What happens with your data, point by point.
- Database and application backend run in Frankfurt.
- For classification we send the feedback text and the project description to OpenAI in the US. Not for training. Screenshots never go to the AI, and neither does the page URL, browser data, sender name or email.
- Screenshots live in private storage, reachable only via time-limited signed links.
- No ad tracking: no ad pixels, no Google Analytics. Fonts self-hosted.
- The widget needs no cookie banner on your site. The visitor session ID is only created on submit.
- Diagnostic capture can be switched off: one attribute on the script tag,
data-diagnostics="off". - Tenant isolation via Postgres row-level security.
- DPA available on request, on every plan.
Founding Offer: 40 spots, one year of Pro at half price. The first 10 start with six months free.
The first ten workspaces get 100 % off Pro for six months and then 50 % off for another six months. The next thirty get 50 % off Pro for twelve months, that is €24.50 instead of €49 per month. Both apply to Pro on monthly billing. We hand out the spots in booking order, redeemable until 31 October 2026. After the twelve months you pay the regular price. You use a young product and tell us what we can add or do better.
Secure one of the 40 Founding Member spots:
Free €0 · Starter €19 · Pro €49, per workspace, not per seat. → All details
Frequently asked questions
When can I start?
Closed beta, we open it up in groups. You will hear from us within a few days.
What does it cost?
Free, Starter 19 €, Pro 49 € per month and workspace. Founding Offer for the first 40 workspaces: the first 10 get Pro with 100 % off for six months and 50 % off for six more, the next 30 get twelve months at 50 % off. We hand out the spots in booking order, redeemable until 31 October 2026.
Do my users need an account?
No. No sign-up, no extension. Name and email are optional.
What exactly gets captured?
Text, screenshot, clicked element, recent console messages, failed requests, page URL, browser and window size. → Privacy
Does the feedback go to an AI?
Text and project description go to OpenAI (US), not for training. Screenshots never.
Can I turn off the diagnostic data?
Yes, with one attribute on the script tag: data-diagnostics="off". Console, errors and network are then never captured; text, screenshot and element marker stay.
Does this work with my stack?
One script tag on any page that serves HTML. Snippets for HTML, Next.js, WordPress, Webflow. It also works inside app shells with a WebView (Capacitor, Cordova, React Native WebView).
What about native iOS or Android apps?
Not yet. The widget is JavaScript and needs a web page; a native SDK is on the roadmap. If your app has a WebView portion, you can embed it there today.
Do I need to update my privacy policy?
Yes, one paragraph. You stay the controller for your users' data, we process it on your behalf. We give you the ready-made policy text and the DPA, both in the project settings.
Where is the data stored?
Database and backend in Frankfurt. DPA available on request, on every plan.
Claude Code, Cursor, Codex?
All three. The setup assistant writes the configuration itself: for Codex in ~/.codex/config.toml, for the others as JSON.
Anything else? Write to us.
Feedback that tells you what to build.
Founding Offer until 31 October: one year of Pro at half price for the first 40, the first 10 start with six months free.
We build Feedbapp with Feedbapp.