Agents for Robotic Compliance

Compliance evidence that keeps pace with every release.

For robot makers, importers and integrators selling in the EU. The agents turn your engineering work (ROS 2 packages, code, tests and releases) into the evidence that notified bodies, customers, market authorities and courts ask for, and keep it current with every release. Agents draft the evidence; your engineers review and approve every item before it counts.

What changes in the EU

Five legal regimes, within six months.

Between September 2026 and February 2027, robots sold in the EU come under new or changed rules for machinery safety, cybersecurity, product liability and batteries. Software becomes part of the regulated product, and every update has to be accounted for.

DateRuleWhat it means for robot makers
11 Sep 2026 Cyber Resilience Act, reporting obligations Actively exploited vulnerabilities and severe incidents are reported through ENISA's Single Reporting Platform: an early warning within 24 hours, a notification within 72 hours, then a final report. This includes products already on the market. Fines reach €15M or 2.5% of worldwide turnover.
9 Dec 2026 Product Liability Directive (EU) 2024/2853 Applies through national law to products placed on the market after this date. Software, AI and updates can count as products. Non-compliance with safety requirements leads to a rebuttable presumption of defect.
20 Jan 2027 Machinery Regulation (EU) 2023/1230 Replaces the Machinery Directive with no grace period: machines placed on the market from this date must comply. Machines and safety components whose safety functions rely on machine learning always need a notified body. New requirements on protection against corruption, importer obligations and ten-year document retention.
18 Feb 2027 Battery Regulation, battery passport A digital passport for industrial batteries above 2 kWh, which covers larger mobile robots.
11 Dec 2027 Cyber Resilience Act, full application Most remaining obligations for connected products apply, including security requirements and conformity assessment.
2 Aug 2028 AI Act, as amended by Regulation (EU) 2026/1744 AI safety requirements for machinery are set through delegated acts under the Machinery Regulation.
In practice

What the new rules ask of a robot that ships software every week.

Updates and the CE marking

A software change the manufacturer did not foresee can be a substantial modification, and whoever makes it takes on the manufacturer's obligations. Changes the documentation describes and limits stay outside that definition.

Proof of the state of the art

Not every harmonised standard will be cited by January 2027, so the presumption of conformity is limited. Where it does not apply, the manufacturer shows compliance with its own evidence.

Machine learning in safety functions

Machines and safety components whose safety functions rely on machine learning are listed in Annex I Part A and always go to a notified body, with an argument for why the function is safe and the evidence behind it.

Importers and procurement

Importers are liable for robots whose technical files are incomplete or only partly translated, and customers' safety officers ask for the risk assessment, the Declaration of Conformity and cybersecurity evidence before they sign.

Compliance pipeline

Every release, checked and documented.

Every release gets one verdict: Ship Review Notified body needed

The agents are installed inside your own infrastructure and work there, next to your Git repositories and CI. Your code never leaves it. A safety.yaml file in each repository declares the safety functions and the code, parameters and model versions that implement them. Agents draft; a named engineer approves every item before it counts.

For the safety lead: every change traced

From the code diff to the measure, hazard, requirement and test, with a check against the documented limits of foreseen modifications.

For the CEO: one verdict per release

What changed, what it affects, what has to happen before you ship, and whether a notified body is needed.

Based on real Nav2 releases: compare 1.3.10 and 1.3.11, or see the collision monitor fix on GitHub.

Specimen for illustration. AMR-1, its safety functions, hazards, tests and limits are fictional, and the vulnerability figure is illustrative. Code excerpts from Nav2 (ros-navigation/navigation2), Apache License 2.0. Independent example, not affiliated with or endorsed by the Nav2 project. Mapping to legal requirements and standard clauses is verified by a safety engineer in every client project.

Partner exchange

One interface for every partner, in both directions.

Integrators, customers, auditors, importers, service partners and insurers connect their own AI agents or tools to your exchange through the Model Context Protocol (MCP). Each partner belongs to a tier with its own permissions, so the same questions no longer arrive by email in different formats.

Partners read approved facts
  • Integrators: speeds in collaborative mode, stopping times, foreseen modifications, residual risks the cell must cover.
  • Customers' safety officers: whether a firmware version is covered by the current declaration, and its security status.
  • Notified-body auditors: the evidence for a given requirement, with time-limited access.
  • Importers and distributors: whether the file is complete, and in which languages.
  • Service technicians and insurers: allowed firmware versions, safety checks, certification status.
Partners send data in
  • Component suppliers: safety parameters, certificates, declarations of incorporation, battery passport data.
  • Software suppliers: SBOMs and vulnerability notices, which start the CRA workflow when needed.
  • Integrators and operators: site modifications, checked against the foreseen limits first.
  • Operators and service partners: field incidents and near misses, for CRA reporting and the risk assessment.
  • Auditors: review findings, which open tasks against the technical file.
A continuous service

Requirements that stay current with the law.

Regulation keeps moving after the application dates: harmonised standards are cited, delegated acts follow, and member states add national rules. We keep the requirements in your safety graph current with EU and national legislation, and the agents show which releases and documents each change affects.

  • Requirements updated as EU regulations, harmonised standards and delegated acts change.
  • National legislation of the member states where you sell, included in the same model.
  • Every change traced to the affected requirements, evidence and releases.
  • Documents and partner answers in English and all other official EU languages.
What you receive

Documents your auditors, customers and lawyers can use.

DocumentWhat it containsUsed by
Living risk assessmentHazards, risks, mitigations and test evidence, linked to codeSafety lead, notified body
Requirements matrixEach legal requirement, how it is met, the evidence and the standard clauseNotified body, market authority
Technical file per releaseThe full dossier, rebuilt with every releaseNotified body, court
Declaration of ConformityThe legal declaration, drafted and ready to signManaging director
Release impact reportShip, review or notified body needed, with reasonsCTO, release manager
Foreseen-modification limitsDocumented limits that keep updates and site changes from becoming substantial modificationsProduct, legal, integrators
SBOM and vulnerability reportSoftware components and known vulnerabilities per releaseSecurity lead, customers' IT security
CRA incident packPre-filled reports for ENISASecurity lead
Importer verification reportWhether a non-EU maker's file is complete and credibleImporters and distributors
Evidence archiveA tamper-evident record kept for ten yearsLegal, insurer
How your knowledge stays protected
  • The agents run inside your own infrastructure. Your source code never leaves it.
  • The partner MCP server runs in a separate environment that holds only the approved facts. It has no physical access to your code or internal data.
  • A fact is disclosed only after your safety lead approves it for a partner tier, release by release.
  • Every answer carries the release version, document ID, hash and signature, and "not disclosed" where a fact is not approved.
  • Incoming data is validated, signed by the sender and reviewed before it is used.
  • OAuth 2.1 with short-lived tokens and enterprise SSO, rate limits and a full query log.
What stays with your engineers
  • Approval of every hazard entry and requirement mapping the agents draft.
  • The decision to ship a release.
  • Signing the Declaration of Conformity. Certification stays with the notified body.
Works with

The engineering tools you already use.

Self-hosted by default: the agents run inside your own infrastructure, so your code stays where it is. As an option, we host the partner exchange and the evidence archive in the EU; only approved facts and signed documents go there. Approved facts are also published as Asset Administration Shell (AAS) submodels for your industrial customers' systems.

  • GitHub
  • GitLab
  • CI pipelines
  • Jira
  • Linear
  • ROS 2
  • Container images
  • Test and simulation reports
  • Model registry
  • CycloneDX SBOM
  • Asset Administration Shell
Getting started

From a readiness check to every release.

1

Readiness check: we review your safety file and release process against the 2026–2027 requirements.

2

Setup: safety.yaml, the CI connection and the first technical file draft, in your infrastructure.

3

Every release: checked, documented and signed, with requirements kept current as the law changes.

Make every release audit-ready