TL;DR
EuConform is an open-source, browser-first tool that helps teams map AI systems to the EU AI Act, run language-model bias tests, and produce Annex IV–style technical reports. The app runs entirely client-side for privacy, offers WCAG 2.2 AA accessibility, and includes an explicit legal disclaimer that it provides technical guidance—not legal advice.
What happened
A developer published EuConform, an open-source application aimed at helping implementers and developers assess parts of the EU AI Act. The tool packages an interactive risk-classification quiz based on Article 5, Article 6 and Annex III, a bias-detection workflow using the CrowS-Pairs methodology, and an in-browser PDF generator structured to mirror Annex IV technical documentation. All processing is designed to run locally in the browser via transformers.js with WebGPU, with optional integration with local model hosts (Ollama) for log-probability-based bias metrics. The project emphasizes privacy (no tracking, no external fonts), multilingual UI (English and German), and keyboard-friendly accessibility compliant with WCAG 2.2 AA. The repository includes installation steps (Node.js ≥ 18, pnpm ≥ 10 recommended), tests, CI configuration, and a dual license (MIT and EUPL-1.2). A legal disclaimer in the project notes it is not a substitute for formal conformity assessments or legal counsel.
Why it matters
- Provides a practical, technical way to screen AI systems against specific EU AI Act provisions without sending data to servers.
- Client-side processing and a privacy-first design reduce exposure of potentially sensitive inputs during testing.
- Built-in bias testing and Annex IV–style reporting help teams produce reproducible artifacts tied to the regulation’s documentation requirements.
- Accessibility and multilingual support lower barriers for evaluators across EU contexts.
Key facts
- Project is published on GitHub under the repository Hiepler/EuConform.
- Risk classification implements Article 5 (prohibited) and Article 6 plus Annex III (high-risk) logic from the EU AI Act.
- Bias detection uses the CrowS-Pairs dataset and computes a log-probability based metric; thresholds >0.1 and >0.3 denote light and strong bias.
- All inference and report generation can run 100% offline in the browser via transformers.js with WebGPU.
- Optional local-model support via Ollama is documented; recommended models include Llama 3.2+ and Mistral variants for logprobs.
- PDF report generation is designed to produce documentation aligned with Annex IV technical documentation structure.
- Accessibility goals claim WCAG 2.2 AA compliance and full keyboard navigation.
- Repository lists prerequisites (Node.js ≥ 18, pnpm ≥ 10 recommended) and developer scripts for testing and linting.
- Dual-licensed under MIT and EUPL-1.2.
- The project includes a legal disclaimer stating it does not replace notified-body conformity assessments or legal advice.
What to watch next
- Project activity and releases on the GitHub repository (issues, pull requests, release notes).
- not confirmed in the source
- not confirmed in the source
Quick glossary
- EU AI Act: A European Union regulation establishing rules for the development, placement on the market, and use of artificial intelligence systems within the EU.
- Annex IV technical documentation: A structured set of documentation elements referenced in the EU AI Act that developers must provide for certain AI systems to demonstrate compliance.
- CrowS-Pairs: A benchmarking dataset and methodology originally developed to measure social stereotypes and bias in language models using sentence pairs.
- transformers.js: A library that enables running transformer-based machine learning models directly in web browsers, often leveraging GPU capabilities.
- Ollama: A local model hosting environment referenced in the project for running and serving language models on a developer’s machine.
Reader FAQ
Is this tool legally binding for EU AI Act compliance?
No. The repository includes a legal disclaimer: it offers technical guidance only and does not substitute for formal conformity assessments or legal advice.
Does my data leave my browser when using EuConform?
The project is described as fully client-side and privacy-first, stating no tracking and that processing occurs in-browser.
Which AI models work best with the bias detection feature?
The source recommends using Ollama with models that support log-probabilities, citing Llama 3.2+ and Mistral variants for best accuracy.
Can I use EuConform for commercial purposes?
not confirmed in the source
EuConform 🇪🇺 Open-Source EU AI Act Compliance Tool Classify risk levels • Detect algorithmic bias • Generate compliance reports 100% offline • GDPR-by-design • WCAG 2.2 AA accessible Important Legal…
Sources
- Show HN: EuConform – Offline-first EU AI Act compliance tool (open source)
- Show HN: EuConform – Offline-first EU AI Act compliance …
- EU AI Act Compliance Checker | EU Artificial Intelligence Act
- What Open Source Developers Need to Know about the …
Related posts
- Grok’s new image-editing feature fuels surge of sexualized nonconsensual deepfakes
- CES 2026 worst in show: AI girlfriends, voice‑locked fridges and other flops
- Democrats Ask Apple and Google to Pull X Over Grok’s Nonconsensual Deepfakes