Now in public beta — free while in beta

Your coding agent checked it once. FusionTest checks it on every deploy.

Describe what your app should do in plain English. FusionTest remembers what worked, replays it on every deploy without AI calls, and tells you what broke and since which deploy — confirmed by a separate check, with evidence. No source access needed.

Join the Free Beta See How It Works

goals:
  - name: "Login flow works"
    url: https://app.example.com/login
    goal: "Enter credentials and verify the dashboard loads"

×Checkout completesREGRESSION · last worked a1b2c3d

Failed again in a fresh session · screenshot + steps to reproduce

Built by the team behindWysLeap•PulsairSocial•FusionLeap Digital

WHY FUSIONTEST

What’s still checking after the coding session ends

A coding agent checks one change, once, while you watch. FusionTest keeps checking every flow on every deploy, and a separate system confirms what it finds.

Regression memory

Passing goals are recorded and replayed on every deploy without AI calls, with nobody watching. When one breaks, you see the last deploy where it worked and the first where it didn't.

An independent verifier

The agent that acts doesn't judge. A separate check decides pass or fail from what's on screen, and a bug is only reported once it happens again in a fresh session, with a screenshot and steps to reproduce.

No source access needed

Give it a URL. FusionTest reads the page the way a screen reader does, so it works on staging, on apps from other teams, and on apps you don't own.

Web, mobile and desktop

Web and Electron through Playwright, Android and iOS through Appium, and native Windows and macOS apps. The same plain-English goals on each.

Honest CI gating

Separate exit codes for pass, fail and couldn't-evaluate, so an AI outage never blocks a merge as a false failure. GitHub Actions annotations, Slack and webhooks.

History and flaky-goal tracking

Every run, step and verdict stays on the dashboard. A goal that fails, then passes when retried in a fresh session, is marked flaky instead of being quietly passed.

HOW IT WORKS

Describe it once. Checked on every deploy.

STEP 1

Describe what should work

Write each goal in plain English, against a URL. No selectors and no access to your source code, so it works the same on your app, a staging build, or an app you don’t own.

- name: "Checkout flow"
  url: https://store.example.com
  goal: "Add item to cart and complete purchase"

→ Navigating to page
→ Clicking "Add to Cart"
→ Waiting for cart modal
→ Proceeding to checkout
✓ Verifier: "Order confirmed" on screen
  Recorded for replay

STEP 2

The first run is checked twice

An AI agent works out how to reach the goal, like a real user would. A separate verifier then decides from what’s on screen whether it really succeeded. Only a confirmed pass is recorded.

STEP 3

Every deploy is watched

From then on, each deploy replays your recorded goals in CI without AI calls, with nobody watching. A goal that breaks is retried in a fresh session before it’s reported, so you get a confirmed regression with the last deploy where it worked. If the AI can’t evaluate a goal, that’s reported as an error, never as a false failure.

deploy f3e9a01
✓ Login flow      replayed, no AI calls
✓ Search          replayed, no AI calls
× Checkout flow   regression
  Last worked: a1b2c3d
  Failed again in a fresh session

RECORD & REPLAY

Record once. Replay free.

Passing goals are recorded. FusionTest keeps the steps that worked and the text that proved the goal passed.
Repeat runs replay without AI calls, checking the same proof on screen. Fast, cheap, and the same every time.
When your UI changes, the replay stops matching, the agent works the goal out again from your plain-English description, and the recording is repaired.
Still honest: the agent re-checks each recorded goal every 10th replay, so a recording can’t quietly drift.

One goal on fusionleap.io, measured:

AI agent

134 s

25.9k tokens

Replay

14 s

0 tokens

Recordings are shared across the dashboard, your laptop and CI when the CLI has your API key, so a goal that passed anywhere replays everywhere.

Where FusionTest fits

	FusionTest	Coding agent's check	Scripted E2E
When it runs	Every deploy, unattended	Once, while you watch	Every deploy
Cost of a repeat run	Replayed, no AI calls	Full AI cost again	Free, but you maintain the scripts
Who confirms a pass	A separate verifier, with evidence	The agent that made the change	Assertions you wrote
Needs your source code	No, just a URL	Yes	Usually
When the UI changes	Repairs the recording from your goal	Starts from scratch	Selectors break
CI result	Pass, fail or couldn't-evaluate	None	Pass or fail
Flaky results	Retried fresh, marked flaky	Not tracked	Retries hide them
Mobile and desktop apps	Same goals	Limited	Separate tools

Frequently asked questions

Let your agents ship. Let FusionTest remember.

Record your key flows once and have every deploy checked by a system that didn’t write the code. Free during the beta, no credit card.

Join the Free Beta