Tests that write and heal themselves.
Describe a user journey in plain English. Dexter runs it in a real browser, repairs it when your UI changes, and blocks deploys that break. Then it turns the same journeys into live demos and checks your launch messaging.
Runs on your private AI model · Built on Playwright
Works with any web app, in any stack
- Playwright
- GitHub Actions
- GitLab CI
- React · Vue · Angular
- REST APIs
- Ollama
From URL to a guarded release in an afternoon
- 01
Point Dexter at your product
Paste a URL. Dexter checks it's reachable and drafts a smoke test, an AI first-visit check and a two-minute demo.
- 02
Describe what users should be able to do
“A new visitor signs up and lands on the checklist.” Dexter works out the clicks, then proposes replay steps you accept in one click.
- 03
Gate every deploy
One line in CI runs the smoke suite on each release. When the UI moves, Dexter repairs the step and tells you; when the app breaks, the build stops.
Illustrative output.
End-to-end tests without the upkeep
Write the journey once, in plain English. Dexter drives a real browser like your users do, and it repairs the steps when your UI changes instead of failing on a renamed button.
- Replays fast and deterministically on Playwright; AI steps in only when the UI moved
- Browserless API checks: status, JSON and response time in seconds, no browser needed
- Explore mode hunts for bugs from a goal, with screenshots, traces and video as evidence
- A smoke gate for CI: exit code 1 when something real breaks
Demos that run themselves, live
Write chapters the way you'd brief a colleague. Dexter presents them in a real browser with captions, records the session, and takes requests from the audience while it runs.
- Reuse your tests as demo chapters: the demo never drifts from the product
- Presenter console: “show how pricing works” and Dexter does it live
- Every demo is recorded to MP4 for follow-ups
- Clearly labelled as AI-presented, so audiences always know
Check your launch before you pay for it
Compare your messages and price points on a synthetic panel sampled from your target market. See which message wins, what people object to, and which benefits never land.
- Personas sampled in code from real market data, never invented by the AI
- Every message compared with a control on the same panel, with 95% intervals
- Objections and messaging gaps grouped by meaning; a between-subject price test
- A confidence score that lists every reason for doubt. Directional by design: confirm winners with real users
Keep Playwright. Lose the maintenance.
Playwright is excellent, and Dexter runs on it. Dexter adds everything you'd otherwise build and babysit yourself.
| Playwright on its own | With Dexter | |
|---|---|---|
| Write a test | TypeScript and selectors | Plain English, or record it once |
| When the UI changes | Fix the selectors (or run its AI healer with your own model) | Dexter repairs the step, tells you, and proposes the update |
| API checks | Separate tooling | Same project, seconds per run, no browser |
| Find bugs nobody scripted | — | Explore mode, with screenshots, traces and video |
| Demos and launch checks | — | Built from the same journeys |
| Where the AI runs | Bring your own | Your private model; nothing sent to AI vendors |
Your product never leaves your control
Dexter was built for teams whose legal department says no to sending product data to AI vendors.
Private model
Every AI step runs on your own model server. Your product, code and data are never sent to OpenAI or any other AI vendor.
Proof of ownership
Dexter only drives domains you've verified with a file or a meta tag, so nobody can point it at a site they don't own.
Allow-listed browsing
Each environment lists its domains; Dexter's browser can't navigate anywhere else.
Production is protected
Tests that change data, and AI exploration, never run against production environments.
Secrets stay in a vault
Test logins live server-side and are redacted from logs, reports and AI history.
Honest AI output
Synthetic personas are labelled as simulated; their quotes are never testimonials.
Free while we build it with you
Early-access teams use Dexter at no cost. Prices are published when the public beta opens; early teams keep a founding discount.
Free
For trying Dexter on one product
- Testing with a monthly run allowance
- Browserless API checks
- One demo
Team
For product teams shipping weekly
- Unlimited projects and seats
- CI gate and recorded demos
- Launch Eval studies
Business
For teams with several products
- Explore mode and higher limits
- Domain verification and audit log
- Priority queue
Self-host
For regulated teams
- Dexter and the model on your servers
- No data leaves your network
- Help with setup
Questions teams ask first
Is Dexter built on Playwright?
Yes. Deterministic replay runs on Playwright, and Dexter adds the parts around it: plain-English authoring, AI repair when the UI moves, exploration that finds bugs, API checks, demos and launch checks, all on a private model.
What does “private model” mean?
Dexter's AI runs on a model server you control (we use Ollama). Prompts, screenshots and page content never go to OpenAI, Anthropic or any other AI provider.
Will AI make my tests flaky?
No. Tests replay deterministic steps first. The AI only steps in when a step can't find its element, and Dexter marks that run “self-healed” so you can accept the new steps. A real failure still fails.
How accurate is Launch Eval?
It's for ranking options and finding objections, not for predicting sales. The method behind it reached about 90% of human test-retest reliability for ranking purchase intent in published research, but synthetic panels run optimistic and are weak on small segments. Every report shows intervals and a confidence score; confirm winners with real users.
What can Dexter test?
Any web app or site you own, in any stack, plus its HTTP APIs. Logins use test accounts kept in a server-side vault.
How much does it cost?
Early-access teams use Dexter free while we set it up with them. Paid plans are published when the public beta opens.
Run Dexter on your product
We're onboarding a small group of design partners: startups and product teams shipping every week. You get Dexter free, a setup call, and a direct line to the people building it.
- A smoke suite running in your CI within a day
- Your first live demo, recorded
- A Launch Eval study on your next message or price change