Guardia finds the obligations hiding in your source — undisclosed AI, missing record-keeping, no human oversight — on every pull request. It cites the article, writes the patch, and opens the PR. Your code never leaves your runner.
No account required · Get results in 2 minutes
These tools are completely free and need no registration.
Get your EU AI Act risk score in 2 minutes. Includes a downloadable report.
Try it free →▶️See the full dashboard walkthrough with sample data — no login needed.
Try it free →📚Plain-language breakdown of every article that affects your AI system.
Try it free →💡FAQs, integration guides, and compliance checklists — all public.
Try it free →Feb 2025
Prohibited AI practices banned
Aug 2025
General-purpose AI rules apply
Aug 2026
Transparency obligations (Art. 50) apply
Dec 2027
Full high-risk (Annex III) obligations
Non-compliance fines: up to €35 million or 7% of global revenue.
From risk assessment to regulator-ready exports, all in one platform.
Automatically discover every AI system in your codebase — OpenAI, Anthropic, HuggingFace, LangChain and 50+ more. Produces your Article 11 AI System Registry in seconds.
Ten rules read your source — Python, JavaScript, TypeScript — and trace whether anything a user can reach ends up calling a model. Undisclosed AI (Art. 50) and missing record-keeping (Art. 12) can fail a build. The Article 5 signals never can: they ask a question, because a prohibition turns on purpose and context that no scan can see.
Guardia writes the patch, re-parses the file to confirm it still compiles, and opens the PR. The codemods are deterministic — no model writes your code. Where the right fix depends on something the scanner cannot see, no patch is offered and the finding says why.
A GitHub Action on the Marketplace and a GitLab component in the CI/CD Catalog. Your source code never leaves the runner — the analyzer parses locally and never calls a model.
Instantly know if your AI is high-risk under EU AI Act Annex III. Get clear rationale, article references, and confidence scores.
Auto-generate technical documentation. AI-assisted drafts, PDF/DOCX export, ready for auditors.
Every date that binds you, tracked separately: Article 50 transparency on 2 Aug 2026, high-risk obligations on 2 Dec 2027 after the Digital Omnibus deferral, Annex I products on 2 Aug 2028. Slack and email alerts when new Commission guidance lands.
Answer a few questions about your use case. Get instant risk classification with EU AI Act references.
Upload predictions for bias analysis. AI-assisted Annex IV documentation generates in seconds.
Download compliance packs (PDF, DOCX, ZIP). Share read-only regulator view. Done.
Start free. Upgrade when you need more.
For startups and small teams
For SMBs shipping AI in production
For mid-market and regulated industries
The product is live and you can start today — this is not a waitlist. Leave your email and we'll send founder pricing and tell you when rules for your stack ship.
No spam. Unsubscribe anytime.
Proof
Guardia scans Guardia on every push to both branches. Here is the finding open against us right now — not a sample, not a mock-up. It is on our own production chat endpoint.
The endpoint POST reaches a model via openai.chat.completions.create (an SDK call, in the handler itself). No disclosure was found in this file.
Article 50(1): “Providers shall ensure that AI systems intended to interact directly with natural persons are designed and developed in such a way that the natural persons concerned are informed that they are interacting with an AI system…”
(rule not yet reviewed by legal counsel)
That endpoint streams its response, so there is no object to add a field to — where the disclosure belongs depends on how the interface renders the stream. Guardia offers no patch rather than guessing. A wrong patch in a compliance tool is worse than none.
On 27 July it found nine real defects in our own product in a single day — including three rules that flagged their own definitions. None was reachable by our test suite. All were fixed and shipped the same day, and the write-up is public.
Articles 5, 10, 12, 14, 50
326 backend, 31 frontend
open-source corpus scanned
published, not hidden
Including what it does not establish: no lawyer has reviewed our rules yet.
We launched in 2026 and are onboarding our first cohort of pilot companies. No fake logos, no invented quotes — when our first customers go live, their real results will be right here. Until then, judge us by the scan above and by everything else we publish about ourselves.
Everything you need to know about Guardia AI and EU AI Act compliance.
Start with our free tier. No credit card required.