What if QA tests themselves?
The problem nobody talks about
Every team says "we do QA." But what most teams actually do is write scripts that break on every UI refactor, maintain selectors that rot over time, and spend Fridays debugging why the test suite failed — not the feature.
QA tooling hasn't caught up with the AI era. While code generation, review, and documentation have been transformed by LLMs, testing is still mostly manual script maintenance dressed up in fancy dashboards.
So we built Vectra QA
Vectra is an open-source, agentic quality assurance framework. Instead of writing tests, you deploy agents.
Here's how it works:
A Test Manager agent dynamically spawns specialized sub-agents — UI Explorers that hunt for broken flows, Data Validators that intercept network traffic and verify payloads
Each agent has its own behavioral DNA, memory, and objectives
Agents don't write JSON blobs. They read and write Markdown files in an Obsidian vault — structured with YAML frontmatter, connected with wiki-links
When their mission is complete, they terminate — no persistent daemons, no wasted compute
Why Markdown?
Because test history should be a knowledge graph, not a database dump. Your QA results are browsable by humans, queryable by LLMs, and portable to any tool. No vendor lock-in. Your vault is just a folder of .md files.
What's under the hood
Multi-agent orchestration with dynamic spawning
Playwright for real browser automation
MCP skill system for DOM manipulation, network interception, and DB validation
RAG integration — agents retrieve user stories from vector stores to inform test strategies
Dark-mode Command Center with live SSE updates
Supports OpenAI, Anthropic, MiniMax, Kimi, or local Ollama
Try it
Copy
git clone https://github.com/Artflarex-Lim...
cd vectra-qa
docker compose up --build
open http://localhost:3000
Or run it headless — the agents write directly to the vault.
We're fully open-source. Star the repo, try it, break it, and tell us what you want to see next. Feedback is the only metric we're optimizing for right now.
Replies