Data stays put
QA work runs on your PC or on the company systems you configure. QALab does not need a cloud account to process it.
QALab
One person, one Windows PC. Ollama is the default. Settings can save OpenAI, Anthropic, or a compatible endpoint. Jira, TestRail, and Bugsnag connect from that machine.
QA work runs on your PC or on the company systems you configure. QALab does not need a cloud account to process it.
The current desktop workspace is a commercial product. Releases already distributed under MIT keep that grant.
Jira, TestRail, Bugsnag, Google Drive, and Google Sheets. The packaged API answers only this PC.
FEATURES
Tools grouped the way the app groups them: ask, analyze, generate, report, Arion, connections, and system.
A chat. A greeting answers at once. A message sent during an answer waits in line. It does not invent bugs, cases, coverage, or regressions, and it does not approve a release.
KPIs from Jira and TestRail. A 7-day or 30-day window is a window: a source that did not answer is not scored as zero.
The only capture analysis. It checks copyright and the language you choose. An unread frame stays unread. Video Analysis is gone.
Audits Jira tickets for missing fields and vague text. A batch survives leaving the page. A ticket the source did not return is not scored as zero.
Recommends from the whole project, a Jira fix version, or a custom field. It can name regressions when that reading exists. It does not approve the release.
Draft and refine Game Design Documents with the saved model.
Generate test cases from a GDD or a feature description.
Score cases or sweep a TestRail suite. A Google Sheet sync does not overwrite a TestRail case.
Google Drive is the connection. A sheet is imported from Test Cases with the Google file picker.
Bugs live inside Arion. Drafts, triage, and export to Jira or Linear. Missing values show as unread or a dash, never as 0.
ARION
The execution side of the desk. It appears in the sidebar while the Settings flag is on. The rest of the desk keeps working when the flag is off.
In 0.6.54: Stop and Pause reach the case that is running. A second run cannot take a device another run holds. A case that crashes is its own error and the rest of the run still runs. A build that did not install ends the run before anything runs. A suite verdict comes from the model's pass/fail answer.
A suite or a free exploration for the minutes you chose. A run that cannot start says why. Cases that had not started are marked Skipped.
Its own browser, a saved sign-in, and crash evidence. It can open a saved page without a build.
Android over ADB. The window no longer freezes while the emulator is booting. A refused read names its reason instead of raw JSON.
History on this PC. Executions and results can be deleted. A run left open when QALab quit is closed as an error. A run that checked nothing has no verdict.
SYSTEM
Account, Connections, AI, and This PC. Desktop setup downloads one local model.
An API error never carries a key, token, cookie, or URL query. Diagnostics export redacts tokens and shows the saved file path.
Much more of the interface is in Spanish. Long reads, scans, and uploads have a clock and Stop, and they survive leaving the page.
ARCHITECTURE
Electron + dashboard + local API + Ollama. The packaged API answers only this PC.
START
NSIS setup or portable build for 64-bit Windows 10/11. The installer is unsigned.
If Ollama is already installed, QALab starts it. A slow answer does not reinstall it. 8 GB of RAM is recommended.
Jira, TestRail, Bugsnag, or a Google Sheet. Then ask from the evidence you already have.
Setup and portable builds are published on alkalo/QALabReleases. An update that does not finish keeps the previous app and its shortcuts.