Fully on-premises · your data never leaves
The EU AI Act and the GDPR - every article, recital, annex and definition - rebuilt as typed knowledge graphs, one per regulation. Obligations bound to actors, paragraphs and evidence. Pick a regulation, ask a question, get an answer with citations - never a guess. All of it running inside your own infrastructure.
Every answer is a walk through the regulation itself - node to node, paragraph to paragraph, with the path shown.
The EU AI Act, Decoded - from a wiki you can read to a graph an agent can reason over Read the write-up in AI Advances →We rank job applicants with an ML model. Where do we stand?
Each obligation in the Act is bound to one actor and one paragraph, with the evidence that demonstrates it. A duty with no mapped control isn't a footnote - it's a hole in your graph you can see.
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.
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
Recital 61
Every classification entry point - Annex III use cases, GDPR processing contexts - is tagged by sector. Filter on yours and the graph collapses to the articles, duties and deadlines that actually reach you.
Nine screens, one graph underneath - from the reasoning agent to the spec linter, the raw graph and the model controls. Click any screen to enlarge.
Compliance work touches your most sensitive material - model cards, incident logs, DPIAs, contracts. ACTLAS is installed inside your perimeter, so none of it is ever handed to a vendor's cloud.
Graph store, reasoning engine, document vault and - if you want it - the model itself all run on your hardware or in your private cloud. No telemetry, no phone-home, no shadow copy of your evidence on someone else's infrastructure. Air-gapped installs are supported.
ACTLAS is model-agnostic. Point it at your own account with any major provider - your keys, your contracts, your billing - or keep everything local by serving open models through Ollama. Same graph, same citations, same answers. Route per task if you want: a hosted model for drafting, a local one for anything that touches client evidence.
Each framework is a graph in the same typed language, so a control you evidence once can answer across all of them.
Regulations move: consolidated texts, delegated acts, guidance, standard revisions. Each graph in ACTLAS carries a version, pinned to the source text it was built from, and every change ships as a diff you can review - so an answer you gave last quarter can still be reproduced against the text that applied then.
eu-ai-act v2.4
gdpr v1.9
diffed · changelogged
Compliance surprises are expensive because they arrive late. ACTLAS doesn't wait for a text to be voted: proposals, committee reports and amendments are modelled exactly like binding law - as their own versioned graph, flagged pre-enforcement so a draft duty can never be quoted as if it already applied. Point a system you run today, or a product you haven't built yet, at a draft and see what would attach: which obligations reach you, which of your controls already answer them, what has to change, and how long you would have. When the text is adopted the draft graph becomes the live one - and the analysis you already did carries over.
modelled in ACTLAS from the first draft
Bring one AI system or one processing activity you run today. We'll walk its path through the right framework - classification, duties, gaps - in under twenty minutes, on an instance that lives on your side of the firewall.
Request a walkthrough→No deck. No form. A live graph and your use case.
A product by Adopt Intelligence