AI-powered testing for web apps.
Give Dagonet a URL. It explores your app in a browser the way your users do, learns how it works, and from then on checks every pull request for the things that broke. No SDK, no test framework, and no access to your code.
- build — passed 2m 04s
- unit tests — passed 1m 12s
- Dagonet · 12 user flows — failed 8m 38s 1 regressed · 10 passed · 1 not verified
Submitting a shipping address now sends the customer to an error page. It worked on main three days ago.
Point it at a URL. Get a working test suite.
It explores.
You give it a URL, and a test login if part of your app sits behind sign-in. Agents then drive a real browser against your app — as a first-time visitor, as a signed-in user, as an admin. They click, type, navigate, and recover from dead ends, taking different routes on different runs.
It remembers.
What it learns becomes a picture of your app that grows with every run: the kinds of page it has, the journeys users take through it, the controls on each screen, and which of them each kind of user can actually reach.
It checks every pull request.
When a PR opens, Dagonet reads its description — the same words your reviewers read — and uses what it already knows about your app to decide which flows are worth checking. It runs them against your deployed app and reports back on the PR.
- the kinds of page it has
- the user flows
- the controls on each screen
- which of them each kind of user can reach
The product is the proof.
You never wrote this test. You can still read it.
Dagonet writes tests from what it observed, in plain language — the steps a person would take and what should happen at the end. No test framework to learn, nothing to maintain, and nothing added to your repository.
- 1Add an item to the cart
- 2Go to checkout
- 3Fill in the shipping address and continue
The comment arrives before the reviewer does.
Every pull request gets the flows that matter to it, run against your deployed app. When one breaks, the PR carries what broke, what it did instead, and when it last worked — so whoever opened it can act without opening another tool.
Submit shipping address now fails.
Step 3 of 4 — filling in the shipping address and continuing sent the customer to an error page instead of the order confirmation.
Last passed on main · a3f2d19 · 3 days ago
When it can't check something, it says so.
An automated tester that guesses is worse than none. When Dagonet cannot reach what a test needs, it reports that it could not verify — never a pass — and tells you what would make it checkable next time.
- Why
- No pending invite existed on the target, so there was nothing to accept.
- To make it checkable
- Seed one pending invite for the test account before the run.
never reported as a pass
Other things Dagonet handles while it's there.
Reads the app as it renders.
The agent doesn't work from a static map. It sees the same DOM your users see, in the same browser, on every run.
Continuous, not on-demand.
Daily, hourly, on every deploy. Dagonet runs in the background and only interrupts you when something changes.
Every link, every page, every run.
404s, 500s, redirect loops, broken anchors. Tracked across runs so you can see the day a link started rotting.
AI ships code faster than teams can review it.
Cursor, Claude Code, Lovable and Bolt push features live before a human reads the diff. The QA pipeline built around weekly releases doesn't fit a deploy-on-every-commit world.
AI tools confidently emit code that compiles, runs, and contains bugs. Vibe-coding platforms compound it — code reaches users without anyone reading it.
mabl, Tricentis and similar platforms are priced for enterprises with dedicated QA teams. Most teams shipping AI-generated code do not have those resources.
Cursor, Claude Code, Lovable, Bolt.new and Bubble move applications to production in hours. Manual QA is now the bottleneck — and the existing tools are not priced for the teams who need them.
Built for teams shipping faster than they can test.
Agency heads & technical directors.
Delivering client web applications under fixed-price contracts. Catch regressions before the client does, without staffing a QA team for every project.
Startup CTOs & technical founders.
Shipping fast with three-person teams. The agent gives you the QA pass an enterprise QA team would do — without the headcount or the enterprise price tag.
Product & QA leads.
Drowning in manual regression. Reclaim the hours spent re-clicking the same flows every release and re-route them into exploratory work that only a human can do.
Sign up and join the waitlist.
Dagonet is a Cloud SaaS. Active early-access users get test runs free while we calibrate the agent against real targets and harden the platform.
Paid plans launch at general availability; pricing will be announced ahead of that date. Users on the early-access programme keep their free-tier benefits when paid plans open.
Join the waitlistBuilt by an engineer who got tired of manual QA.
Bryn is the technical founder of Dagonet. His background spans software engineering and applied AI — enough years writing tests by hand to know which parts of QA an autonomous agent can actually take over, and which parts a human still needs to judge.
Dagonet is operated by Anbaric Ideas Pty Ltd, registered in New South Wales, Australia. The platform is deployed continuously from a single mainline branch.