mend.

MEND AI RECOVERY, BUILT INTO YOUR APP

Keep your users
moving forward.

Mend runs inside your app. It detects when a transition takes too long, lets Mend AI choose a recovery you’ve approved, and applies it for the user—then checks that the next screen is ready. Mend AI learns from browser-verified fixes.

Try the free demo

No signup. No card. No cloud tokens.

Developer SDK pilot AI setup required App Store release pending

A little room to recover
An olive-green lamp lighting a quiet workspace
Good UI deserves a safety net.Observe. Recover. Verify.
  1. 01 Detect a slow transition
  2. 02 Select a recovery with Mend AI
  3. 03 Verify the next screen

RECOVER THE USER’S FLOW, IN THE APP

From a stalled transition to a working screen.

Developers integrate the SDK, set the time budget, and register recovery actions. Mend watches each instrumented transition.

Detect the slow or stuck transition

When the next screen misses its deadline, Mend records the failure and requests a diagnosis through your app’s configured backend.

Let Mend AI choose an approved recovery

Mend AI selects from the recovery actions you have registered. It answers on its own once your app’s verified fixes make it confident. Mend uses your approved actions; it does not run code generated by a model.

Recover for the user, then verify

Mend applies the selected action and measures again. If it cannot verify recovery, it reports the failure for developer attention.

MEND AI

Train. Fix. Iterate.

Mend AI is the small model on each Mend server. It gets better at your app’s transitions as the browser verifies more recoveries, and it never runs code a model wrote.

  1. TrainVerified outcomes teach Mend AI

    Each recovery the browser verifies, or sees fail, teaches Mend AI on your server which approved action works for that kind of stuck screen.

  2. FixWhen it is unsure, its reasoning engine corrects it

    Mend AI asks for help only when it is not confident, and keeps each checked answer as a correction. Once the browser verifies enough recoveries for that kind of stuck screen, Mend AI answers on its own.

  3. IterateA nightly benchmark improves the rules

    An automated job tests Mend AI on synthetic test transitions and proposes rule changes for review. It never uses data from your app or server.

TRANSITION HOTSPOTS FOR DEVELOPERS

See where transitions fail in your code, and whether to resolve or redirect.

Mend groups failed and slow transitions by the source label you write on each trigger with data-mend-source, such as src/cart/CartButton.tsx:42, and ranks them by failures and recency. For each hotspot it says where to act, which strategy is safe and how to set it up, with a snippet to copy. Fixed rules decide, so hotspots use no AI. Mend never discovers labels on its own: only the labels you write are used, and data-mend-hotspots="off" keeps them out.

Resolve

Fix it in place

The destination exists but is hidden, late or animating too long. Mend runs the recovery you registered for it—reveal the destination, reduce motion or retry the same navigation—and then measures the screen again.

Redirect

Send the user to your fallback

The destination never mounts, or fixing it in place keeps failing. Once the signal latency has passed, Mend sends the user to a fallback screen on the same site that you registered, and verifies that screen instead.

Review

Leave it to your code

Neither is safe, such as motion that never ends or a screen that keeps changing. Mend reports the hotspot with the most likely cause and a snippet to log its timeline, and does not act.

EXAMPLERedirectRank 1 · confidence 78%

src/cart/CartButton.tsx:42 · Open cart

Failures
7 in the last week: destination missing 6, destination hidden 1
Timing
Median 1,004 ms against a 1,000 ms signal latency (Balanced in 5)
Act on
The navigation, with a redirect to a registered fallback

Why

  • 6 of 7 failures left the destination unmounted at the deadline.
  • No fallback destination is registered yet.
  • No recovery has been browser-verified here yet.

How

  1. Open src/cart/CartButton.tsx:42.
  2. Register a same-origin fallback for cart-view with a navigate(path) callback for your router.
  3. Mend redirects only after the signal latency, then verifies that #cart-help is visible.
  4. Fix why the route does not mount the cart: redirecting is the safety net, not the fix.
redirects: {
  'cart-view': { to: '/cart/help', target: 'cart-help' },
},
navigate: (path, { signal }) => {
  if (!signal.aborted) router.push(path);
},

An illustrative example, not data from a real app. The developer workspace and the companion app show live hotspots from your own Mend server; the free demo below shows the hotspot for each seeded problem.

FREE INTERACTIVE DEMO

See it break. Help it mend.

Choose a seeded problem and a signal latency—how long Mend waits before it signals a slow or stuck screen. Try it, apply the suggestion, and inspect the measurements from your browser.

Loading the local demo… If it does not appear, reload this page with JavaScript enabled.

These are isolated examples, not a repair of your app. The demo’s observations and your signal latency choice stay in this page’s memory and clear on refresh. No account, API calls, or model training.

CLEAR BOUNDARIES

Install it where the transition happens.

The current SDK pilot works in instrumented web apps. Developers connect their AI backend and approve the available recoveries. Mend works within that integration; it cannot repair unrelated installed apps.

Read the AI and data disclosure
CURRENT STATUS

Try the demo today.

The developer pilot is in progress. An App Store release is pending. Online checkout opens soon.

See proposed pricing Existing developer workspace

Owner sign-in only, with an operator-issued gateway login. There is no public sign-up.