Free
no card$0
- Analyses
- 20 / mo
- Fixes
- 5 / mo
For Playwright Test · Node.js 20.11+
Know why the test failed. Fix it only when the evidence says so.
KwikQA reads a failed Playwright run (the JSON report, traces, and test source) and tells you the cause, citing the evidence behind it. When the test itself is at fault, it proposes a small repair, shows you the exact diff, and keeps it only if your tests pass.
npm i -D kwikqaFree plan: 20 analyses and 5 fixes a month. No card needed.
$ npx playwright test ✘ [chromium] › checkout.spec.ts:18:5 › applies a discount code 1 failed · 41 passed $ npx kwikqa analyze Diagnosis request -> the KwikQA service, which forwards it to OpenRouter Model: anthropic/claude-haiku-5.5 via anthropic, effort max Evidence: 12 redacted items from report, trace, source Payload: 4f9c2a7e1b0d83c5... (exactly this payload is sent) Send this request? yes #1 The spec clicks test id "apply-coupon"; the trace shows data-testid="apply-discount", role button "Apply discount". owner: test-support, evidence: strong Result: diagnosed (test-support-owned) $ npx kwikqa fix --- a/tests/checkout.spec.ts +++ b/tests/checkout.spec.ts - await page.getByTestId("apply-coupon").click(); + await page.getByRole("button", { name: "Apply discount" }).click(); Apply this patch? yes Result: awaiting-validation (patch-applied) $ npx kwikqa validate Result: repaired (validation-passed)
After a red run
Three commands: analyze, fix, validate
Each one stops at a clear result, so you always know what happened and what comes next.
-
npx kwikqa analyzeDiagnoses the failures in your latest report. Every cause cites the report fields, trace events, or source lines that support it, and KwikQA says so when the evidence isn't enough.
diagnosedinconclusive -
npx kwikqa fixWhen the cause belongs to the test, proposes a repair. You see the intent it preserves, the exact diff, and the validation commands before you approve it.
awaiting-validationabstained-from-repair -
npx kwikqa validateReruns the failing test in a fresh process. A pass keeps the patch. Anything else rolls back exactly what KwikQA changed. One retry, then it stops.
repairedrepair-failed
Repair policy
Rules every repair has to pass
A green build is easy to fake. KwikQA checks each proposal against these rules before you see it, and rejects the ones that break them.
Never weakens a test
Edits that skip or focus tests, add fixed waits or retries, accept broad snapshots, or swallow errors are rejected.
test.skip(waitForTimeout(retries:toMatchSnapshot(catch (e) {}
Keeps every assertion
A repair that removes an assertion, or uses any matcher fewer times in a file than before, is rejected. KwikQA can change how a test reaches the behavior, never what it accepts.
matcher count ≥ before
Edits test code only
Only the paths your reviewed kwikqa.config.json allows. Never application code, dependencies, lockfiles, or generated files.
specpage-objectfixturetest-datatest-helper
Small by default
A proposal touches at most 3 files and 100 changed lines. Anything larger comes back as a manual recommendation instead of a truncated patch.
≤ 3 files≤ 100 lines
Nothing happens without you
Each AI request shows its model, evidence, cost cap, and payload hash before it is sent. Applying a patch always asks, even with --yes.
Apply this patch?
Leaves your work alone
Files you have already modified are off limits, and rollback restores only the bytes KwikQA wrote.
modified files excluded
What leaves your machine
A redacted, bounded evidence packet, and nothing else
KwikQA builds the packet locally from selected report fields, trace events, and short source excerpts. Secrets and sensitive fields are removed before you see the preview. Screenshots and videos are described by size only; their pixels are never sent, and neither are full files or your repository.
Free, Pro, and Team
your machine→KwikQA service→OpenRouter→model
- The service checks the model route and output cap, counts the request, and forwards the payload unchanged.
- It keeps account, usage, and cost records plus a hash of each request. Never the evidence, your source, or the model's answer.
BYOK and BYOK Team
your machine→OpenRouter→model
- Requests go straight to OpenRouter with your own
OPENROUTER_API_KEY. - Your evidence never passes through KwikQA's servers; the CLI reports only each request's kind and model, to count it.
Pricing
Plans for one developer or a whole team
When you reach a limit, KwikQA stops before sending anything and tells you. Local-only analysis is always free.
Pro
$20 / mo
- Analyses
- 90 / mo
- Fixes
- 20 / mo
Team
$25 / seat
- Analyses
- 100 / seat
- Fixes
- 25 / seat
BYOK
$10 / mo
- Analyses
- 500 / mo
- Fixes
- 150 / mo
BYOK Team
$15 / seat
- Analyses
- Unlimited
- Fixes
- Unlimited
Team plans start at 3 seats and share one monthly pool. Annual billing gives you two months free. Compare plans
Questions
Before you install
Which projects does KwikQA support?
Node-based Playwright Test projects written in TypeScript or JavaScript, in a git repository, on macOS, Linux, or Windows, with dependencies already installed through npm, pnpm, or Yarn. Playwright 1.55 is verified; other versions run with a warning.
What if the failure is a real bug in my app?
Then KwikQA says so. A repair needs positive evidence that the fault belongs to test code. When the evidence points at the application or the environment, you get the diagnosis and where to look, and no patch.
Can it run in CI?
analyze can, with --yes and a token from your dashboard. fix cannot, because a person approves every patch. See Running in CI.
Which AI models does it use?
Hosted plans use Claude Haiku 5.5 for diagnosis and GPT-6 Luna for repair proposals, through OpenRouter. On BYOK plans you can choose your own.
What counts toward my monthly limits?
One analysis is one diagnosis sent to the AI. One fix is one repair attempt, validated by npx kwikqa validate, and its one retry is included. Requests you decline, local-only analysis, and requests that fail before an answer arrives don't count. Plans and usage has the details.
Run it on your next red build
Install, sign in with GitHub, and analyze the failures you already have.