Start here
The AI Act in plain words
…
Not sure where you land? Answer four questions → · See the full risk pyramid →
Myth-busting
Five things people get wrong about the AI Act
The Act attracts a lot of hearsay. Here's what it actually says — each answer links to the provision it rests on.
Live status
Where the AI Act stands today
Today
…Applies now
Coming up
National context
And in Slovakia?
The AI Act is an EU regulation — it applies directly in Slovakia.
Already binding here
…
The national law (in progress)
…
Quick check
60 seconds: where do you fall?
Two questions, then a tailored starting point that opens into the full guide. A screening aid, not legal advice.
The guide
Does it apply to me — and what exactly do I have to do?
Four questions decide which regime you fall under. Then pick your role and follow the exact path the Act itself prescribes — every step is a real provision you can read in one click.
What's at stake
What a breach can cost
Fines scale with the violation and your worldwide turnover — a fixed ceiling or a percentage, whichever is higher (for SMEs, whichever is lower).
Illustrative maxima under Art 99 and Art 101 — not legal or financial advice.
The Act itself
One law, five regimes, 119 articles
The pyramid is the Act's logic: the rarer the use-case, the harder the rules. The spine below is every article in order — click anything to read it in English and Slovak.
↑ stricter rules, fewer systems · ↓ lighter rules, most systems
Plain-language glossary
The jargon, in one line each
The terms the Act keeps using — tap any to see what it actually means, then jump to the provision.
Use it in chat
Ask the Act itself — in your own AI, at no cost to anyone
This site runs a free MCP server. Connect it to Claude, ChatGPT or any MCP client and your AI answers with exact citations, current deadlines and both language versions. Your model, your account — the server only serves the law.
Connect this MCP server to Claude, ChatGPT or Cursor in one step — full setup and all 16 tools on the connect page.
Audit
Point it at your project, get an AI Act gap report
Run an AI coding agent (e.g. Claude Code) inside your repository with this server connected, invoke the audit prompt, and it will inventory your AI capabilities, classify their risk, and check the key obligations for your role against your code and docs — with file-level evidence.
Inventory
The agent scans your repo for AI capabilities: models, APIs, automated decisions, generated content.
Classify
Each capability is screened against Art 5, Annex III and GPAI rules — verified against the actual provision text.
Gap-check
Obligations for your role are checked as covered / partial / missing, citing your files as evidence.
Deadlines
Every finding carries its date — binding law today vs. post-Omnibus — so you know what's urgent.
The slash command works in an interactive session. For CI or headless runs, pull the prompt via prompts/get and pass it as text.
Going deeper: obligations about data (Art 10, 12, 19) can only be verified against what is actually stored — file sizes, ages, and whether any job deletes them. With read-only file tools the audit cannot see that and will mark those rows [unverified]. Two ways to close them. Preferred, and the right default for a device holding personal data: let the audit tell you which read-only commands to run, run them yourself and paste the output back — the agent never gets elevated rights. Or grant a shell allowlist. Note honestly what that costs: a plain Bash(sudo ls:*) does not work, because globs expand as the unprivileged user, so the agent needs Bash(sudo bash -c:*) — and that pattern matches any command, so it is effectively unrestricted root. In that mode the read-only property rests on the prompt and on your approval gate, not on the allowlist.
Technical screening with citations — not legal advice. The report is a head start for your lawyer, not a replacement. The audit prompt never modifies files, creates or deletes anything, or executes your code. With read-only file tools that is enforced by the toolset itself; if you grant a shell, it is enforced by your client's permissions and the prompt — so keep the allowlist read-only.
Open & free
Nothing to pay, nothing to trust blindly
Bring your own model
The server never calls an LLM. It serves the law — search, texts, deadlines, checklists. Intelligence comes from your own AI subscription, so hosting stays free.
Verbatim EUR-Lex source
Texts come straight from the EU Publications Office (CELLAR), parsed with stable per-article identifiers. A daily watcher flips the Omnibus dates the moment they become law.
MCP server
16 tools over the law — search, provisions, deadlines, risk classification, penalties. Connect it to your own AI and ask in plain language.
Setup + all 16 tools →