AI compliance record · EEA

"Which AI systems do we run,and who approved them?"

Someone will ask. The board, a customer, an auditor, counsel. Andes keeps the answer current: every AI system, its candidate EU AI Act obligations with the article cited, and the evidence. One governed record instead of nine spreadsheets.

Pinned EU AI Act referenceArticle-level provenanceHuman review gatesEvidence and deadline tracking
Synthetic UI study · not product evidenceBuyer hierarchy preview
KKernalIntelligenceCommercialExecutionComplianceGraph
AI system inventory3 systems
Recruitment assistant
Deployer · HR · Review in progress
Support triage
Deployer · Customer service · Owner assigned
Document classifier
Provider context unresolved · Legal operations

Recruitment assistant

Candidate screening support · EEA workforce

Review in progress
RoleDeployer
Intended useDecision support
OwnerPeople operations
Candidate obligation pathsCitation-backed
Article 26
Deployer obligationsRecorded role and intended use support this candidate path. Human review required.
Review
Article 27
Fundamental rights impact assessmentApplicability depends on deployment context and unresolved organisational facts.
Unresolved
Source
Authentic regulation textArticle-level citation, reference release, and facts used remain visible.
Pinned
The scattered version of this answer is where you are today. This is the record.

The gap

AI adoption moves faster than the record around it.

A model ships inside a vendor tool. A team quietly adopts an agent. A pilot changes scope and nobody renames it. Six months later the board, a customer, or an auditor asks the question in the headline, and the answer is a two-week archaeology project. The consultant's audit doesn't save you: it was true for the week it was written.

A policy says what should happen. A governed record shows what was actually reviewed, decided, and maintained, and stays true next quarter. Andes is the second thing.

The cockpit

One record. Four jobs.

01

Inventory the systems

Every AI system your organisation runs, with its intended use, role, risk context, owner, and review state. If it's not in the inventory, it's not governed.

02

See why a rule may apply

Recorded facts connect to candidate EU AI Act obligations with the article cited and the open questions kept visible. You can check every step.

03

Assign the work

Evidence, owners, review status, and deadlines stay attached to the obligation they answer. The decision never drifts away from its record.

04

Keep the history

When people, vendors, or models change, the facts, citations, and decisions that explain your current position are still there.

Show the path

Every explanation is checkable.

If you don't trust AI vendors on compliance: good. Neither do we. That's why Kernal shows its work: the recorded facts it used, the article it cites, the path it traversed, and what it still does not know. Uncertainty is displayed, never papered over with a confident answer.

Control before commitment

Reads never change records. Previews are zero-write. When a mapping is ambiguous, the system stops and asks a named human. That refusal is a feature. It is exactly what a reviewer wants to find.

What is live

Built for reviewable evidence, with the boundaries visible.

Reference

Pinned and read-only

The EU AI Act reference cartridge is mechanically read-only, with its release identity displayed.

Provenance

Article-level sources

Authentic regulation citations stay attached to every candidate obligation path.

Review

Ambiguity stops

One-to-many mappings wait for named review. Nothing links durably on a guess.

Access

Environment scoped

Credentials and routing deny cross-environment access, stated exactly as proven and no further.

And what we do not claim

We keep a public list of the things this product does not do: no compliance guarantees, no legal advice, no claims ahead of validation. In this category, that list is the credential. Read it.

Underneath the cockpit

Kernal keeps the evidence connected.

Kernal is the governed evidence graph under Andes. It holds each AI system together with the facts, obligations, citations, evidence, owners, decisions, and deadlines that explain its compliance work, as one connected record rather than a folder of exports.

The same engine supports broader institutional memory and agent context. Compliance is the first public operating problem, not the limit of the platform.

Working session

Bring one AI system. Leave with a clearer map.

Thirty minutes. We map the system, the facts that shape its obligation path, the evidence you already hold, and the questions that still need review. The answer becomes a record you open, not a project you launch. You keep the map either way.

Reference and operational support only

Not legal advice or a compliance determination.