When you add AI to a collaborative robot cell, you don't just add AI. You add the execution gating, the deterministic safety boundary, and the traceable logging the EU Machinery Regulation's new AI-safety review expects. You add the documentation for an ISO 10218-2 risk assessment.
You build it on one project, ship it, and build it again on the next. That's the engineering cost AI-enabled integration carries right now — and it doesn't show up as billable work.
The integrator legally owns the liability for the complete AI-enabled cell. The compliance burden lands on you, not on the AI vendor.
xZane is a hardware-agnostic AI execution runtime — the orchestrator and the enforcer. It routes AI commands, loads device schemas, builds a unified capability surface, and holds the deterministic line between what the model intends and what the device is allowed to do. It never touches hardware directly; it always goes through a HAL.
You write two things: a HAL adapter that speaks your hardware's language, and a capability schema that declares what your device is allowed to do. The schema is the containment mechanism — the model can only request what the schema explicitly sanctions.
Everything else — the enforcement engine, the audit logging, the reflex safety watchdog — is xZane. Write the adapter once. It's there on every project after.
In a traditional robot cell, the arm and hardware are only about 25–40% of total project cost.1 The majority goes to integration — engineering, programming, glue code, hardcoded safety logic and commissioning — which on its own routinely runs 30–50% of the project.2 A custom AI-enabled integration commonly takes 8 to 26 weeks.3 That's before the AI model changes, or the client adds another robot.
xZane's architecture eliminates the majority of that integration work at the source. Here's exactly where those hours go.
These are actual xZane runtime responses from a ROS 2 robot. The same runtime architecture, unchanged, running across three different hardware stacks.
// AI asked to drive forward for 2 hours
{
"status": "rejected",
"error_code": "parameter_out_of_bounds",
"violations": [
"duration_ms=7200000 exceeds max 20000"
],
"intent": "robot.move.forward",
"ts": "2026-08-12T20:03:52Z"
}// Legal command — stopped by the world
{
"status": "interrupted",
"outcome": "interrupted_by_reflex",
"reason": "reflex_latched_mid_motion",
"latch_reason": "front",
"ai_command": null,
"confidence": null
}Demonstrated across ROS 2 robotics, consumer appliances, and smart-home devices — same runtime architecture, three hardware stacks.
A robot cell is machinery, so its operative EU law is the Machinery Regulation (EU) 2023/1230 — mandatory from 20 January 2027, with no grace period.5 For the first time it requires a safety review of AI and self-evolving behaviour, a documented risk assessment of the integrated cell (ISO 10218-2), and traceable technical documentation. xZane produces the logging and the documented safety envelope this demands as a byproduct of normal operation.
Under the 2026 Digital Omnibus reform, AI-enabled machinery is governed by the Machinery Regulation rather than the EU AI Act's Chapter III directly, with AI-specific machinery requirements due by August 2028.6 Where a system does fall in the AI Act's own scope, the same audit log maps to its Article 12 record-keeping requirement — one layer covers both.
Two deadlines are already on the clock: the Machinery Regulation applies 20 January 2027 (no grace period), and AI-specific machinery requirements are due August 2028. Every AI-enabled cell you ship between now and then needs this layer.
One device, one use case, 4–6 weeks. We write the HAL adapter together. At the end, xZane is running live on your hardware — and your engineers understand every part of it, ready to reuse it across hardware variants.
xZane was personally designed and built by Julio Chinchilla — the runtime, the validation gate, the independent reflex mechanism, and the hardware abstraction layer. The work spans runtimes, data platforms, and embedded systems, built between Vilnius and Tallinn.
A US non-provisional patent is filed. The SDK is ready to hand to a first integrator.
Tell us a little about your device and what you'd want a pilot to prove. A 30-minute call is usually enough to scope whether the pilot makes sense for your project.
Or write directly to [email protected]
Or email [email protected]
Cost and timeline figures are industry ranges and vary by application; regulatory positions reflect the state of EU law as of 2026 and are not legal advice.