ACTLAS rebuilds the EU AI Act, the GDPR, NIS2 and DORA as typed knowledge graphs, with a reasoning agent on top. Ask what applies to your system and it walks the actual law - and answers with citations. Map your specification onto the graph and the duties nothing addresses become visible. Record what you actually do against each one. And keep all of it on your side of the firewall - a laptop is enough.
What you'll open every morning
Everything above - the nodes, the edges, the typed structure - exists so that one window can be trusted: a chat. Describe the system you run in plain language and the agent classifies it against the law, binds the obligations that follow, and answers with the sources attached.
It is not a chatbot with legal flavour. It never answers from memory - every claim is retrieved by walking the graph, the walk is drawn on screen while it happens, and every retrieval is logged. Checking an answer takes a click, not an afternoon.
Classification first, then every duty that attaches - deployer and provider separated, cited article by article, in language you can forward.
The traversal panel is the sub-graph behind this exact answer, drawn live while the agent works - nodes and edges of the actual law, not an illustration.
Every retrieval logged with its trust label - verbatim legal text or curated mapping - so “where does that come from?” is answered before anyone asks.
In every company shipping AI into the EU, someone owns an uncomfortable question: which of these obligations actually reach us - and what can we show for them? Their options today are a stack of PDFs, a €15,000 one-off assessment that starts aging the day it lands, or a €50,000-a-year governance suite that wants the company's data in someone else's cloud.
ACTLAS is the fourth option. It treats a regulation as what it actually is - a network of duties, actors and definitions - and holds everything a regulator will ever ask you on three connected layers:
Every article, annex, definition and duty as a node; every cross-reference a real, labelled edge. Verbatim, versioned, never modified.
Your specification, mapped onto the duties it touches. A duty nothing maps to is a visible gap - not a footnote in a report.
The logs you keep, the human who reviews, the assessment you ran - recorded against the exact duty each one answers to. Append-only and hash-chained: an altered record breaks the chain, and the tool names where it broke.
Who gets what out of it
All of it runs where your data already lives: one small machine inside your perimeter - a laptop is enough - with your own model key, or fully local models through Ollama. Nothing you upload, ask or record ever has a way out. That isn't a deployment option. It's the design.
Thirty-nine seconds of the real thing: a system specification overlaid on the EU AI Act, coverage gaps in the linter report, and the evidence layer lighting up over the base graph. No cuts, no mockups.
You may know our open Obsidian vaults on GitHub - that's the readable edition. The graphs inside ACTLAS are the working one: typed, directed relations with a trust label on every edge, and actor, sector, citation-path and version metadata on every node - 620 nodes and 3,279 edges for the Act alone. The reasoning engine walks that structure with the path shown live. Neither can drift from the text, because neither ever leaves it.
None of the method is secret. We published how the Act was taken apart and rebuilt, edge by edge - read it and judge the modelling for yourself →
We rank job applicants with an ML model. Where do we stand?
Every enumerated point carries a block anchor from its citation path. When an auditor asks “where does it say that?” - you don't scroll, you link.
Article 14 ^p4
Annex III ^5-c
GDPR Article 33 ^p1
The agent classifies, retrieves, and explains what must be true - and refuses to rubber-stamp. The conclusion stays where regulators expect it: with you, on evidence you can point to.
Governance suites start at €50,000 a year because they carry a cloud, a sales team and your data. ACTLAS carries none of those.
Founding partners pay the rates below, keep 25% off list for the life of the subscription, and get the evidence ledger free for the first year. Standard rates are printed on every card. Licences run 3, 6 or 12 months at the same monthly rate; founding pricing needs the 12-month term and a shorter term is credited in full if you switch while a slot is open.
Claim a founding slot→For one person
€10/ month
foundingbilled €120 a year
Standard €25/month · €300 a year
For one company
€100/ month
foundingbilled €1,200 a year
Standard €200/month · €2,400 a year
For a firm serving clients
€300/ month
foundingbilled €3,600 a year
Standard €600/month · €7,200 a year
For a group
from €25k/ year
billed annually
Scoped by entities, deployments and frameworks
Add-on module
Record what actually happens - which model ran, which articles it consulted, who claimed what - filed against the exact duty it answers to, graded by how well it is evidenced, append-only and hash-chained. It is the part that turns a gap analysis into something you can show an auditor.
Free in the first year for founding partners. Included with Enterprise. Priced at half the base licence on every tier, so it scales the way your licence does.
Your data exports at any time - it is in your database on your hardware, and everything exports to JSON or CSV. Add-ons: extra deployment +50% · air-gapped delivery €2,500 a year · extra client workspace €480.
Behind ACTLAS is a network of AI governance, compliance and engineering specialists we bring in when a deployment needs shaping - a regulation nobody has modelled, tuning to how your sector works, an integration into the stack you already run.
It does not have to stay a separate application either. ACTLAS can be installed as an engine inside your existing systems - the graph, the reasoning and the evidence layer reachable from your own portal, workflow or product, with the interface you already have. Tell us where the gap is. It is usually a scoping call, not a roadmap request.
Prices exclude VAT · terms apply. Every tier is set up together with the Adopt Intelligence team, and the graphs, updates and change reports are included throughout. Support beyond the included hours is €120 an hour, or €1,000 for a ten-hour pack. Free trial on request - a guided sandbox with demo data, nothing to install.
No GPU, no cluster, no vendor cloud. The only cost that scales is the model you choose - and you pay that provider directly, or run models locally and pay nobody.
2 vCPU, 8 GB RAM. The laptop you already own, spare VM capacity at €0 marginal cost, or roughly €45-90 a month on a small Azure instance. No telemetry, no phone-home. Air-gapped installs supported.
Bring your own key - your account, your contracts, your billing. Or bring no key at all and serve open models through Ollama. Same graph, same citations, same answers.
One AI system you run today, twenty minutes, a guided sandbox. You'll see where it lands on the Act, which duties attach, and what's missing - live, on the graph, nothing to install. If it fits, your own instance goes up behind your firewall the same week.
Book the twenty minutes→No deck. No form. Founding 30 open - from €120 for the first year.
A product by Adopt Intelligence