vegaml WCAG 2.2 detector

Paste HTML markup and score it against all 63 WCAG 2.2 success criteria with vegaml (frozen LLM reader + 14.3M decision engine). The probabilities are vegaml's calibrated outputs. The reader sees markup text only: contrast- and layout-only criteria need computed styles to be conclusive.

Model: Turbolego/vegaml-wcag-embed-v1 · Repo: github.com/turbolego/vegaml-wcag

Server

No model loaded yet.
Backend / connection notes →
This page is served via HTTPS, so it can only reach HTTPS backends (browsers block mixed content: HTTPS page → http:// fetch). The default backend is the hosted vegaml-serve on Modal (hummernkloa--api.us-east.modal.run) — it scales to zero, so the first call after idle takes ~1 min to wake.
  • Default (hosted, no setup): just click Check server — it wakes and answers; or override with ?apiBase=<url>
  • Own machine: run vegaml-serve --model 0.8b --host 0.0.0.0 --port 8080 --preload --cors and open the local mode (python3 server.py → http://127.0.0.1:8090, API base stays /api)
  • Faster GPU backend: add gpu="A10G" to the Modal specs in deploy/modal/app.py (the 63-question batch drops from minutes to seconds)

HTML to check