← All Products & Workloads / Artel QA
EXPERIMENTAL · QA HARNESS Experimental · Lite on npm

Artel QA

Reads the source, drives a real browser, rules from evidence.

Our experimental QA harness that maps every interactive control directly from your source code, exercises your running application in a real browser, and adjudicates pass/fail verdicts strictly from network, DOM, and database evidence. Artel QA Lite runs locally on your own agent via npm.

Join Early Access · 45% Off → See Pain, Solution & Proof ↓
Artel QA
Source-Aware Browser QA Harness
1. Identify (Source) Reads route and component source code to list every control and its intended side effect.
2. Walk (Zero-Cost) Opens pages and confirms which controls exist live before burning model tokens.
3. Inspect (Browser) Exercises controls in a real fenced browser while recording network and DOM traces.
4. Adjudicate (Proof) The harness writes the verdict from recorded measurements, attaching the proof.
5 / 5

Seeded broken controls caught with full network & screenshot proof

0 / 4

False alarms on healthy controls during benchmark verification

$0.36

Total agent cost for a full verification run on your own subscription

npm

Artel QA Lite runs locally on your machine and your own coding agent

VALUE ADDITION · ARTEL QA

The Pain Point, Our Solution, and Why Metantel.

The Pain Point

AI-generated code ships faster than humans can test, and raw browser MCPs guess

When coding agents ship features in minutes, manual QA becomes the bottleneck. Meanwhile, giving an agent a raw browser MCP (like Playwright MCP) leaves it guessing: it doesn’t know what a button is supposed to write to the backend—so if a UI toast says "Saved!" while the network request silently failed, a raw browser tool marks it passed.

The Solution

A white-box QA Harness: source inventory + deterministic evidence adjudication

Connect a QA Harness, not a raw browser MCP. Artel QA reads your source code first to inventory every control and its expected handler behavior. It then drives a real browser, captures actual HTTP requests, DOM state, and screenshots, and uses deterministic harness rules—not LLM vibes—to issue the verdict.

Why Metantel?

Zero-config verification that feeds directly back into Flow Console and matures into Agentify

Artel QA caught 5 of 5 broken controls with 0 false positives for $0.36 on our seeded test suite. Connected to EDIP, it tests against your team’s product specs; once adopted on a web app, the exact map it builds matures into Agentify’s governed agent APIs.

What you gain with Artel QA by role
For Individual Developer

Run Artel QA Lite via npm on your own machine and Claude subscription before opening a PR—catching silent UI-to-API regressions for pennies.

For Engineering Team

Hand stakeholders and reviewers an audited report where every bug has the exact network payload, console trace, and screenshot attached.

For Enterprise

Connect Artel QA to EDIP so tests verify compliance against documented business rules and runbooks—laying the foundation for Agentify.

Why a QA Harness Beats a Raw Browser MCP

A raw browser MCP can see that a button exists on screen and click it. It cannot know that clicking that button is supposed to POST a specific payload, trigger a backend service chain, or mutate a specific database row—unless a harness reads the source code and team specs first.

Artel QA combines white-box code reading with real browser execution. Crucially, the LLM does not get to guess whether a test passed: the Artel QA Harness inspects the captured HTTP traffic, status codes, and DOM mutations against explicit evidence rules.

Five-Stage Verification Pipeline

  • Identify: static source analysis lists every interactive control and page route so coverage gaps are visible immediately.
  • Walk: deterministic browser walk confirms live reachability and authentication before model tokens are spent.
  • Inspect: agent-driven browser interaction exercises forms, modals, and workflows inside a sandboxed environment.
  • Adjudicate: evidence rules verify network requests, response payloads, and UI states against source expectations.
  • Report: generates a stakeholder-ready verification matrix with reproducible steps, network logs, and screenshots.

Artel QA Lite & The Path to Agentify

Lite on npm · Bridges Testing to Production Agent APIs

Artel QA Lite runs locally on your machine using the coding agent you already pay for. And because Artel QA learns every control, form, and permission boundary of your web application during testing, adopting it creates the exact structural map needed for Agentify to expose your web app as a safe, audited agent interface.

UNIFIED ARCHITECTURE

How Artel QA connects to the Metantel stack

Every Metantel product shares one underlying memory and governance layer. Context, verification proofs, and access rules flow between tools automatically.

Artel QA Closes the loop → Flow Console

Sends proven failure traces back to Flow Console coding agents so they fix the exact root cause.

Explore Flow Console →
Artel QA Reads requirements → EDIP

Pulls product requirements and team decisions from EDIP so tests check business intent.

Explore EDIP →
Artel QA Matures into APIs → Agentify

Once Artel QA maps and verifies your web app’s controls, Agentify turns them into governed agent endpoints.

Explore Agentify →
Explore the rest of the Metantel line-up
Flow Console Beta EDIP Beta · Open to Partners Agentify Under Development Artel Workspace Live Beta Elenchus In Development