karoslabs
NightShift QA
The overnight QA agent that only files bugs it can prove. NightShift explores your web app while you sleep and reports a bug only after reproducing it in a fresh browser with the AI completely out of the loop.
Why proof is the product
Private alphaEvery AI testing tool has the same failure mode: it says it found a bug, and half the time it hallucinated. One false report and you stop trusting the tool. NightShift's answer is structural. The AI proposes actions, deterministic code watches for real failures, and a suspected bug counts only when the exact recorded action trace triggers the same failure again in a clean browser session, twice. No replay, no bug report.
How a night runs
-
1. ExploreA Claude brain proposes one action at a time. Real clicks, real forms, a real Chromium browser driven by Playwright.
-
2. DetectDeterministic oracles watch every page: crashes, console errors, failing API calls, dead links. Not the AI's opinion. Hard evidence only.
-
3. ProveEvery suspected bug is replayed from its recorded trace in a fresh browser context with the AI removed. Same failure signature or it is discarded.
-
4. ReportEach confirmed bug ships with numbered repro steps, a screenshot, and a standalone Playwright script that exits 0 when the bug reproduces. Drop it into CI as a regression test.
Bring your own brain
-
Your Claude, your machineNightShift runs entirely on your computer against your app. Use the Claude subscription you already have, or your own API key with one config line. No hosted service, no pooled credentials, no telemetry.
-
Budgeted by defaultConservative limits on calls, minutes, and sessions per night. It works within your account's limits and stops at the hour you set.
Design partners
NightShift is in private alpha and I am taking a small number of design partners: teams shipping fast without a QA person who want overnight coverage with proof. It runs nightly on my own products today. If you want it running on yours, write to me and tell me what you are building.