AI QA Agent
An AI agent that executes your test cases, judges the results, and never pretends.
Reads your test cases in natural language, runs them in a real browser, and tells you the truth about what happened.
What this replaces
Automation has always meant code
Every test case becomes a script that someone has to write and maintain.
Scripts break when the UI changes
QA engineers spend their time fixing selectors instead of testing the product.
AI that reports what it never saw
AI test tools can mark steps as passed that they never actually performed.
What it does
Steps are written in natural language; the agent works out how to perform them.
Reads test cases written in natural language
Import a markdown playbook or a spreadsheet, or write cases directly. Steps, expected results, and dependencies come in intact.
Executes them in a real browser
Drives Playwright through logins, forms, tables, and multi-step flows, recording screenshots, video, and network activity as evidence.
Verifies results, honestly
Every step is checked against its expected result. If it can’t be checked, it’s marked unverified — never green.
Self-heals before it marks a fail
A failing step is diagnosed and, where it can recover, adapted and retried. Recovered steps are marked self-healed, with what changed.
Built to be trusted near real data
Deterministic where it matters
Common actions — clicking, filling forms, reading tables — run as built-in skills, not model guesses. Every run shows how many steps ran deterministically.
A tie is a refusal
If a step says “click Delete” and the page has two, the agent stops and reports both. It never guesses near real data.
Small models, your hardware
Runs on compact open models with a single 16 GB-class GPU, or larger ones if you have them. vLLM, LM Studio, Ollama, or a cloud API — your choice.
Self-hosted, end to end
Deploy on-premise, air-gapped, or in your own cloud. Nothing runs on our infrastructure, and nothing phones home.
Features
- Natural-language test cases with inputs, expected results, captures, and dependencies
- Import from markdown playbooks and a spreadsheet template; AI extraction for other documents
- Real-browser execution with automatic login and session reuse across runs
- Screenshots, video, and network capture as per-step evidence
- Honest verdicts: unverified, needs-review, blocked, and skipped are all distinct from pass and fail
- Self-healing before a step is marked failed, with every healed step flagged and the change shown
- Deterministic skills with per-run coverage reporting
- AI exploration: crawl an application and propose test cases for review
- Knowledge-grounded chat over your own documentation and run history
- Test suites, run history, and analytics
- Jira and GitHub issue filing — off by default, nothing is filed without your say-so
- SSO (SAML), RBAC, audit log, encrypted settings
- On-premise, air-gapped, or your own cloud; bring your own model and embedder
What it connects to
Bring your own model and embedder. Runs wherever you deploy it.
Made for these teams
Test cases, no automation
QA teams who have test cases but no automation.
Automation that breaks
Teams whose automation breaks faster than they can fix it.
Data that can’t leave
Organisations that cannot send application data — or test credentials — to a cloud AI service.
Be first to know
The AI QA Agent is in build and coming soon. Email hello@alnizam.dev to hear more.
Get notified at launch
Same standard, different job
AI SOC Agent
Your SOC, standardized. Your data, at home. Self-hosted security operations for teams without a 24/7 SOC.
AI Data Agent
Put your data strategy into practice. A guided, self-hosted workflow from business question to decision to outcome.
Coming soonQuestions this page did not answer?
Ask us directly.
or email hello@alnizam.dev