Last updated: 27 October 2026
The VerifactAI API classifies a declared AI system against the EU AI Act, returns the obligations that apply to it, enforces the controls that can be enforced at runtime, and appends a tamper-evident evidence record for every check. The n8n node and the Make app are thin clients of this API; everything they do, you can do directly.
Base URL: https://api.verifactai.app
Authentication
Every /v1 request carries an API key in the Authorization header:
Authorization: Bearer ak_live_…
Keys are issued when an account is created, and sent by email. A key gives access to every system of the account. There is no key for a single system.
Language
Text fields (labels, obligation titles, error messages) are returned in English by default. Ask for French with ?lang=fr on the URL, or with the header Accept-Language: fr. Identifiers, codes, dates and option values never change with the language.
Errors
Errors use one shape. The code is stable and meant for branching; the message is for humans and may change.
{
"error": {
"code": "system_not_found",
"message": "System not found",
"details": null
}
}| Status | Code | Meaning |
|---|---|---|
| 400 | invalid_json, invalid_declaration, invalid_attestations, invalid_name, invalid_strict_mode, unknown_pack | The request is malformed. details lists each problem. |
| 401 | missing_api_key, invalid_api_key, revoked_api_key | The key is absent, unknown or revoked. |
| 402 | system_quota_exceeded, feature_not_in_plan, overage_not_available | The plan does not cover the request. details.capacity says what it would cost. |
| 404 | system_not_found, evidence_not_found | The resource does not exist, or belongs to another account. |
Rule packs
GET /v1/packs
Lists the published versions of the rules. latest is the version a new system is pinned to.
{
"latest": "[email protected]",
"packs": [
{ "ref": "[email protected]", "published_at": "2026-08-11", "jurisdiction": "EU", "rules": 16 },
{ "ref": "[email protected]", "published_at": "2026-10-17", "jurisdiction": "EU", "rules": 36 }
]
}GET /v1/packs/{ref}/schema
The declaration form of a version: the facts to declare, the controls that can be implemented, the recognised standards, and the fields accepted in an attestation. Build your forms from this endpoint rather than from a hard-coded list; a new version of the rules changes it.
{
"ref": "[email protected]",
"facts": [ { "id": "role", "type": "enum", "group": "Identity", "label": "Your role under the Regulation", "options": ["provider", "deployer", "importer", "distributor"], "labels": { … } }, … ],
"controls": { "disclose_ai_interaction": { "label": "AI interaction disclosure", "layer": "runtime", "test": "…", "automatable": true }, … },
"standards": [ { "id": "en_18286_2026", "label": "EN 18286:2026 — …", "kind": "harmonised_standard", "presumption": "…", "covers": ["quality_management_system"] } ],
"attestation_fields": { "owner": "string", "document_ref": "string", "verified_at": "date (ISO)", … }
}Controls have a layer: runtime controls can be enforced on every call, config controls are verifiable on the product, org controls are internal processes that no API can enforce.
Systems
A system is a declared AI use: a support chatbot, a CV screening tool, a product-description generator. One workflow may call several models; it is still one system.
POST /v1/systems
Declares a system. The declaration is validated against the pack: an unknown fact or control is rejected, never ignored.
{
"name": "Support chatbot — Acme",
"facts": { "role": "provider", "eu_nexus": true, "interacts_with_humans": true, "generates_content": true },
"implemented_controls": ["disclose_ai_interaction", "interaction_logging"],
"control_attestations": {
"interaction_logging": { "owner": "Head of Engineering", "document_ref": "RUN-012", "verified_at": "2026-10-01", "valid_until": "2027-10-01" }
},
"strict_mode": "dry_run",
"pack_ref": "[email protected]"
}| Field | Required | Notes |
|---|---|---|
name | yes | Shown in the register. |
facts | yes | Keys from schema.facts. role is mandatory. Omitted booleans count as false. |
implemented_controls | no | Control ids from schema.controls. |
control_attestations | no | Object keyed by control id. Fields in schema.attestation_fields. Past valid_until, the control no longer counts as in place. |
strict_mode | no | off (default), dry_run or on. on requires a plan with blocking mode. |
pack_ref | no | Defaults to latest. |
Returns 201 with the system, including its id (sys_…) and the pack_ref it is pinned to.
GET /v1/systems
All systems of the account: { "systems": [ … ] }.
GET /v1/systems/{id}
One system, as declared.
PATCH /v1/systems/{id}
Updates any of name, facts, implemented_controls, control_attestations, strict_mode, strict_policy. Send pack_ref to migrate the system to another published version: the declaration is re-validated against it, and evidence already recorded keeps the version it was produced with. Empty strings are ignored, so a form can send every field.
Checks
POST /v1/check
Evaluates a system and appends an evidence record. Call it before the AI call and branch on decision.
{ "system_id": "sys_8f2a…", "model": "gpt-5", "purpose": "customer_support", "actor": "support-bot", "dry_run": false, "record_evidence": true }Only the system id and this optional metadata are sent. Prompts, responses and message content never transit through the API.
{
"decision": "warn",
"risk_level": "transparency",
"pack_version": "2026.10.1",
"evaluated_at": "2026-10-27T10:00:00.000Z",
"mode": "dry_run",
"obligations": [ { "id": "ART_50_1_DISCLOSURE", "article": "Article 50(1)", "title": "Inform the user they are interacting with an AI", "risk_class": "transparency", "triggers": [ … ] } ],
"upcoming_obligations": [ … ],
"controls_missing": ["disclose_ai_interaction"],
"controls_applied": ["interaction_logging"],
"controls_to_prepare": [],
"expired_attestations": [],
"would_block_in_strict": ["disclose_ai_interaction"],
"policy": { "strict": false, "on_missing": "warn", "layers": ["runtime"], "escalated_by": [], "exempted": [] },
"evidence_id": "evt_m2k9x1c",
"disclaimer": "…"
}| Decision | Meaning |
|---|---|
allow | Every required control is in place. |
warn | A required control is missing. Nothing is blocked. |
review | The system is high-risk and a human decision is expected. |
block | A prohibited practice (Article 5), or a missing runtime control in strict mode. |
controls_to_prepare lists controls required by obligations that are not yet enforceable: the roadmap. expired_attestations lists controls whose attestation has lapsed.
Evidence
GET /v1/systems/{id}/evidence?limit=100
The evidence chain of a system, verified server-side. integrity.valid is false if any record was modified, removed or reordered, with brokenAt and reason.
{ "system_id": "sys_…", "count": 18, "integrity": { "valid": true }, "records": [ { "evidence_id": "evt_…", "sequence": 0, "previous_hash": "0000…", "hash": "3cba…", "pack_version": "2026.10.1", "decision": "warn", "obligations": [ … ], "controls_applied": [ … ], "controls_missing": [ … ], "recorded_at": "…" } ] }Each record hashes identifiers only, so the chain does not depend on the response language.
GET /v1/evidence/{id}
One record, as signed.
Register and dossier
GET /v1/register
The account's register: one line per system with its current decision, risk level, counts of obligations and controls, missing controls, expired attestations and last check, plus a summary.
GET /v1/systems/{id}/detail
The full compliance view of a system: obligations enforceable today and upcoming, controls by layer with their attestations, controls to prepare, applicable standards, chain summary and sources. Available on every plan.
GET /v1/systems/{id}/dossier
The same content as a printable HTML document, in French or English (?lang=fr|en), or as JSON (?format=json). Export requires a paid plan.
Account
GET /v1/entitlements
The plan in force, the number of systems declared, the allowance, and the features enabled.
POST /v1/keys
Issues an additional API key: { "label": "Production", "email": "[email protected]" }. The key is returned once and emailed.
GET /v1/capacity/quote?additional=1 and POST /v1/capacity
Quote, then purchase, additional systems beyond the plan allowance (card subscriptions only).
Versioning
Each system is pinned to the version of the rules it was declared with. A decision taken in March is reproducible with the law of March. New versions are listed by GET /v1/packs; migrate a system with PATCH /v1/systems/{id} and pack_ref.
Disclaimer
The API identifies the obligations applicable to a declared system and records the controls applied. It does not constitute legal advice and does not guarantee compliance.