Site navigation and actions.
Hydrofoil opens each pull request's preview deployment in a real browser, checks the change against your ticket, and posts bugs, accessibility problems and visual issues to GitHub. Every finding comes with a screenshot and repro steps. There are no tests to write or maintain.
Checkout button does nothing on a slow network
src/checkout/PayButton.tsx:42Error message isn't announced to screen readers
src/checkout/AddressForm.tsx:88Mobile layout clips the postcode field
src/checkout/address-form.css:17“Sign in” and “Log in” are both used
src/components/Header.tsx:12Low confidence, so it won't fail the check.
End-to-end suites are powerful, but someone has to write and maintain them, and they only catch the scenarios someone encoded.
No reviewer can realistically click every new flow, resize the window, trigger every error state and read the accessibility tree.
By the time someone finds the broken interaction, the work has moved on and the fix costs more.
Hydrofoil closes the gap between code review and real-world QA.
Hydrofoil is notified through the GitHub App.
The deployment, diff, PR description and linked ticket.
Isolated, with deterministic checks first and AI judgement on top.
A check, a summary and inline comments, each with evidence.
Every finding shows its work.
Every Hydrofoil finding comes with evidence, so your team can see what went wrong and decide what to do about it.
file:line for code findings.Findings without evidence are dropped.
Low-confidence findings are marked Needs review and never fail your check.
Dismiss a finding with a reason and it stays dismissed.
It builds acceptance criteria from your Jira or GitHub issue, the PR description and the diff. Then it verifies each one in the code and in the browser, and flags changes that no ticket explains.
You can edit the criteria. Your edits carry over to later runs.
src/lib/analytics.ts changes tracking. Nothing in the ticket explains it.What happens when…
The API takes five seconds?
the loading state
Pay button shows no loading state for 5s
The request fails?
the error state
A 500 leaves a blank panel
The user double-clicks?
duplicate submissions
One click, two orders created
The user refreshes halfway through?
whether state survives
Step 2 of the form clears on refresh
The window is 390px wide?
clipping and overflow
Postcode field runs off screen
The user only uses a keyboard?
focus and interaction
[TBD: only if live]Focus is lost after closing the dialog
The session expires?
what the user sees
Expired session shows a blank page
“Log in” here and “Sign in” there?
terminology consistency
Two names for the same action
Install the GitHub App. Open or update a pull request. Hydrofoil waits for your preview deployment, runs its reviews, and reports back to GitHub. Nothing to migrate, and nothing to host.
A check run, a summary comment, inline comments on code findings, and merge gating at the severity you choose. Re-run any time with /hydrofoil run.
Vercel, Netlify and Cloudflare Pages, or any per-PR URL pattern.
Jira Cloud and GitHub Issues.
[TBD: Linear, Slack and Figma as “Coming later”?]Hydrofoil only needs read access to your code.
Real applications aren't public demos. Give Hydrofoil test personas and it can use them to review authenticated flows. Credentials are encrypted, never shown to the AI model, and redacted from logs and screenshots.
| You already use | Great at | Can't see | |
|---|---|---|---|
| Playwright or Cypress tests | Repeatable checks on known flows | Flows nobody scripted, copy, design, accessibility | Reviews what the PR touches, with nothing to write |
| An AI code reviewer | Deep reading of the diff | The running app | Adds runtime behaviour |
| Visual regression tools | Pixel-level baselines | Whether a new design meets the ticket | Judges the change against the ticket |
| Manual QA | Human judgement | Being on every PR | Does the repetitive pass; people take the exceptions |
| A coding agent in your editor | Fast, flexible, free | Independence and consistency on every PR | Runs on every PR whether or not anyone remembers |
Credits track real compute. Every run shows an estimate beforehand and the actual cost afterwards, with a breakdown per review. Owners can set monthly and per-run spend caps. Your first PR is free.
See pricingIllustrative numbers.
Install the GitHub App, point it at a preview deployment, and see what it finds. No test scripts to write. No migration.
Review your first PR free