COMPLIPATHDOC complipath.io/homeRENDERED 2026-08-23ENGINE 2026-08-09.1CORPUS 2024/1689 + 2026/1744 + Commission guidelines

Find your AI systems. See which the EU AI Act covers.

Free, and you don’t need an account. You get a real answer from the same engine the product runs, and it shows you which article of the law it came from. Keep the result and it becomes your register.

Example assessment · engine output, fetched at this build

An example system, run through the real engine. Never written by hand, never a mock-up.

Kestrel Applicant Ranking (test)high
point 4(a) of Annex III
8 obligations from 2 December 2027
1 obligation from 2 August 2026

The full run, published in full — same payload, same source, fetched at this build.

First system free

Complipath is EU AI Act compliance software for AI-heavy software companies without a compliance team — an AI system register, deterministic risk classification, the obligations that follow, and the evidence behind every decision.

Complipath is early and says so. Your first system is free — no application, no programme, nothing asked in exchange — and what the product does today is written out on this page, along with what it does not.

What Complipath is, in plain language — who it is for, who it is not for, and what it costs.

Every verdict names the corpus it was derived from, and cites the article it rests on.

“Classification derived from Regulation (EU) 2024/1689, Regulation (EU) 2026/1744.”

What the check cannot decide is published, not hidden.

See a complete run, including what it left undetermined →

Illustration — not an assessment
AI Register 8 systems · 2 high-risk
Lumen Screening
HR / Recruitment
point 4(a) of Annex III High
CreditSense Scoring
Credit & lending
point 5(b) of Annex III High
Support Copilot
Customer support
Article 50(1) Limited
Churn Forecast
Revenue operations
Minimal
Every system carries its classification, its obligations and its evidence in one place. Lumen Screening, CreditSense Scoring, Support Copilot and Churn Forecast are invented systems; the article references show the form a verdict takes, not a verdict on anything.
03 · The engine

What the engine decided, and what it refused to decide

Both halves below come from a single assessment, fetched from the app while this page was built. Nothing on this card is written by hand. An example system, run through the real engine.

Decided
Kestrel Applicant Ranking (test) high
point 4(a) of Annex III

You answered that the system is used for the recruitment or selection of people — in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates.

confirmed 17 August 2026 · engine 2026-08-17.2
Refused — requires legal judgement
Article 6(3)

Article 6(3) can take an Annex III system back out of high-risk. The engine does not apply it, and this is the provision it is declining to reach a conclusion on. It reaches a system only where it “does not pose a significant risk of harm to the health, safety or fundamental rights of natural persons, including by not materially influencing the outcome of decision making”. It applies only “where any of the following conditions is fulfilled”:

  • (a) “the AI system is intended to perform a narrow procedural task”
  • (b) “the AI system is intended to improve the result of a previously completed human activity”
  • (c) “the AI system is intended to detect decision-making patterns or deviations from prior decision-making patterns and is not meant to replace or influence the previously completed human assessment, without proper human review”
  • (d) “the AI system is intended to perform a preparatory task to an assessment relevant for the purposes of the use cases listed in Annex III”

“Notwithstanding the first subparagraph, an AI system referred to in Annex III shall always be considered to be high-risk where the AI system performs profiling of natural persons.”

Each of the four is a judgement about intended purpose, and the profiling limb overrides all of them. This engine reaches none of it. It records that the question is open, names the provision and leaves the conclusion where Article 6(4) puts it: with the provider, who must document that assessment before the system is placed on the market or put into service.

quoted from sources/regulation-2024-1689.html · 2026-08-20
01 · Why now

Three reasons this stops being optional

Article 113

The rules are here

The EU AI Act is in force and its obligations phase in on fixed dates — dates that moved in July 2026. Registration and conformity assessment kept their original date; classification, documentation and human oversight moved to 2 December 2027.

point 4(a) of Annex III

AI is outrunning oversight

Teams ship models faster than anyone writes them down. The screening tool nobody registered is the one a reviewer asks about first.

Article 99

Later costs more

Reconstructing a year of decisions under a deadline costs more than recording them as they happen — before penalties enter the conversation.

From scattered files to one system of record

Before
Compliance lives in five places at once
A spreadsheet of models someone maintained until March
Risk notes in a Notion page, half of them from a workshop
Evidence scattered across Drive, Slack threads and email
No answer to "who owns this system" that two people agree on
A compliance question that stalls the deal for weeks
After
One register, one status, one export
+One register with every system, its owner and its risk level
+A classification you can defend, with the reasoning attached
+Evidence linked to the obligation it satisfies
+A named business owner on every system
+An export you can send the same afternoon it is asked for
02 · The product

What you get, and when

Nothing here is gated to a higher tier. Some of it is not built yet — the list says which.

01
Live
AI Inventory & Register

One list of every system your team runs or builds, with purpose, domain, risk level, owners and status. This is the record a reviewer reads.

02
Live
Risk Classification

A guided assessment that returns a risk level, the provisions it rests on, and the reasoning behind it. You confirm or contest it. What it can and cannot decide.

03
Live
Requirements Checklist

Every obligation the classification triggers, with status, owner, evidence link and notes. This is where the work happens week to week.

04
Live
Documentation Workspace

Overview, data description, risks and mitigations, human oversight — one structure per system, base fields auto-filled from the register record.

05
Planned
AI Drafting Assistant

Drafts each documentation section from what you already recorded. It stays a draft until a named person on your team confirms it.

06
Live
Dashboard

Total systems, high-risk count, open obligations and completion, without assembling a report.

07
Live
Market & Law

What changed since you last logged in, in plain language, with the provision it sits under and which of your systems it touches.

08
Live
Weekly Snapshot by Email

The same summary reaches everyone with access, every Monday, so compliance does not depend on someone remembering to check.

09
Live
Named Owners

A named business owner on every system, so 'who owns this' always has one answer. Add a technical owner when there is one.

10
Live
Export

Take your data out at any time. Two families in three formats each — the whole register, and any single assessment — as CSV, JSON or PDF, plus a PDF of the Annex IV sections from the documentation workspace.

11
Planned
Domain-specific Guidance

Short contextual tips per domain. HR flags bias, fairness and appeal mechanisms; credit flags scoring logic and adverse decisions.

12
Live
Evidence Hints & Inline Help

Each obligation carries guidance on what counts as evidence, with domain-specific help where the terms get technical and inline explanations where they actually exist.

Before you decide
What this check can and cannot decide →

Five things the classification decides, five it hands to a human. Written once, by a person, against what the engine actually covers — never generated per visitor.

Why we say we are early

Compliance work is specific. A register that fits an HR screening tool does not automatically fit a credit scoring model, and the difference shows up in the evidence a reviewer asks for.

So we would rather be told what breaks than guess at it. Classification logic that is wrong about a real register gets rewritten now, while it is one product decision, rather than after it has become a default that thousands of teams inherit.

That is the whole reason this site lists what the product does not do, article by article, next to what it does. Being early is not a thing to survive until launch — it is the part where what you run into still changes what gets built.

Early is a reason to check rather than to believe, so what a stranger can verify about us without asking us is a page of its own.

Complipath is built by Yobel Tzegai in Gothenburg, Sweden. The product records who confirmed every classification, so it should say who built it.

What it costs
Your first system is free
Always, not until a date. There is no application, no programme and nothing asked in exchange — it is what the first system costs.
The rate is held
Anyone who has confirmed a classification on or before 31 December 2026 keeps the Starter rate for twelve months from the day they first pay. That commitment is part of the terms.
The check needs no account at all
The classification runs in your browser before you have signed up for anything, and the reasoning and citations come with it.
What the product does, written out in full
How the four risk outcomes work, article by article What an AI inventory records, and why each entry carries its version of the law What goes in the Annex IV file, and when it must exist
04 · Guides

Written by the person who built the classifier

What counts as an AI system under the EU AI Act?

The qualifying test in Article 3, point (1), limb by limb — including the two limbs that die in every summary — and the separate question the Regulation never answers: how many AI systems you have.

Is your AI system high-risk? A decision tree

Five yes/no questions to classify an AI system under the EU AI Act: the Article 5 gate, both high-risk routes, the Article 6(3) exemption, Article 50.

What is the EU AI Act timeline? Every deadline from 2024 to 2030

Every EU AI Act deadline from entry into force on 1 August 2024 to the 31 December 2030 transitional cut-off, with the article behind each date.

Am I a provider or a deployer under the EU AI Act?

The Article 3 definitions of provider and deployer in plain language, the practical test between them, and the Article 16 vs Article 26 obligations each role carries.

All 34 guides →
05 · Pricing

Pricing

Starter
€499/month
Growth — not available yet
€1,499/month
Scale — not available yet
€4,999/month

Your first system is free. What each plan includes, what is built today and what is not, and the questions people ask at the price — the pricing page has all of it.

Questions

Is Complipath legal advice?
No. Complipath provides legal information, and a documentation workspace you write in, over base fields filled from your register. A named person on your team confirms each section, and that confirmation is what makes it your evidence.
What happens when the law changes?
Guidance, standards and national decisions land in the Today view with a plain-language summary and a note on which of your systems it touches.
Do we need this if none of our systems are high-risk?
The inventory and classification still matter — you need to be able to show how you reached that conclusion, and limited-risk systems carry transparency obligations of their own.
Is it launched, or is there a waiting list?
It is open. The check runs with no account, the first system is free, and Starter can be paid for today. There is no waiting list and nothing to apply for. Growth and Scale are not built yet, and the pricing page says which is which.

Classify your first system

Free, and it stays free. There is nothing to apply for and nothing asked in exchange.

Classify your first system

The check runs in your browser before you sign up for anything. Or write to hello@complipath.io. That is a person’s inbox, not a form.

We measure page views with Vercel Web Analytics. It uses no third-party cookies. Visitors are identified by a hash derived from the incoming request, which is discarded after 24 hours, and no identifier is stored that could follow a visitor to another site. What is collected: the time of the visit, the URL, the referring page, filtered query parameters, city-level location, operating system, browser and device type.