AI-driven QA, without the IDE or terminal
Client
BMINC
Year
2026
Duration
4 months
Platform
Web
Technologies
QA Console is a web platform that lets non-technical QA operators run AI-driven quality assurance against live internal applications. Users describe a test in plain language, pick the repositories and environment, and watch an AI agent drive the app in real time — then receive a graded, searchable Markdown report. All of it happens in the browser, with no IDE, terminal, or developer knowledge required.
A powerful AI QA stack — skills, rules, browser automation, and a strict report format — already existed, but it could only be operated by developers inside an IDE. The QA team had no way to run tests, sync source code, or read results without engineering help, creating a bottleneck and scattering reports across files.
A full-stack control surface that wraps the entire QA stack in a clickable UI: a guided run wizard, a single-flight queue that streams the agent's live log, read-only GitLab source sync, an environment catalog, and a full-text report browser with PASS/FAIL/PARTIAL grading. Secrets are encrypted at rest and the production build ships compiled bytecode only.
QA Console turns a developer-only agent stack into a three-step web workflow any QA operator can run.
Pick a preset or write a prompt in plain language, add optional acceptance criteria, and select the repositories and environment to target.
The run is queued and an AI agent drives the live app while you follow the streaming log. Cancel any time; runs are isolated and time-boxed.
Get a graded Markdown report with findings, repro steps, and suspected source files — fully searchable and exportable to PDF.
Describe a test in plain English; the Cursor agent drives the live app and reports findings. One active run at a time via a single-flight queue.
A 6-step flow — preset, prompt, expected outcome, repos, model, review — guides any QA operator from intent to a queued run with validation at every step.
Optional acceptance criteria turn open-ended runs into graded contract tests: the agent returns PASS, FAIL, or PARTIAL with supporting evidence.
Watch the agent think and act in real time, with cooperative cancel, a wall-clock timeout, and full forensic stream logs for every run.
Every QA session is indexed with full-text search (SQLite FTS5), outcome and severity filters, print-to-PDF, and email delivery.
Register any GitLab repo and run branch-aware, read-only source sync so the agent always reads current code across multiple projects.
Dashboard with run stats and recent activity
Log in to the admin portal, create a new record, edit it, and confirm that validation rejects invalid input and the changes persist.
Guided run wizard: prompt, repos, and model
Live agent log streaming during a run
Searchable reports browser with outcome filters
A background worker spawns the Cursor CLI agent headless and parses its stream-json events into a live log, token usage, and a final graded outcome. Concurrency is controlled by a single-flight queue with a global sync lock so runs and source syncs never collide.
QA reports are Markdown files indexed lazily into SQLite FTS5 for instant full-text search. The browser renders server-side Markdown with outcome and severity filtering, print-to-PDF via PDFKit, and SMTP email delivery with PDF attachments.
A self-hosted, single-VM deployment behind a reverse proxy with internal TLS and systemd-supervised services. Secrets are encrypted at rest with AES-256-GCM, and the production release ships compiled artifacts only — the worker is compiled to V8 bytecode and first-party source is stripped on deploy.
QA runs entirely from the browser — no terminal, no Cursor
Plain-language prompt to a queued, graded agent run
Multiple GitLab projects registered and testable
Full-text search across every QA session ever run
Wherever a team needs repeatable, evidence-backed checks against a live application, QA Console removes the engineering bottleneck.
Re-run preset scenarios against the latest branch after every change.
Quick confidence checks that critical flows still work end to end.
Describe a reported issue and let the agent confirm and document it.
Grade a build against explicit acceptance criteria before release.
Every run is logged with who ran what, the model, and the outcome.
Target several registered repositories and environments from one console.
We build AI-powered platforms that turn complex, developer-only workflows into clean web tools your whole team can use. Let's talk about yours.