01 Accountability
Who answers when the AI acts?
A statistical proposal commits no one. A physical action commits your company, your teams and your insurer. Accountability cannot be delegated to a model.
Physical Decision Intelligence
AI proposes. Who authorises?
DECIVER claims neither to drive the machine nor to certify it. It is the sovereign sub-system that turns an action proposed by AI into a decision that is authorised, constrained, traceable and technically verifiable before it is physically executed.
Intelligences propose without pause. A single layer decides what reaches the machine.
01 The problem
Between “the model suggests” and “the machine acts”, a layer is missing
The AI systems that drive robots, lines and machines grow more capable every month. But a probabilistic proposal is not an industrial decision.
Between “the model suggests” and “the machine acts”, a layer is missing: the one that checks every action against the constraints of safety, physics, operations and regulation — and keeps the proof of it.
02 Auditability
Can you demonstrate why the machine did that?
Six months after an incident, the decision has to be reconstructed: which constraints applied, which data was available, which alternative was set aside and why. An event log is not enough.
03 Trust
How do you deploy more autonomy without losing control?
Without proof, the trade-off collapses into two extremes: restrict the AI until it is useless, or let it run without a net. No serious industrial firm signs off on either.
02 What DECIVER does
It stands between the intelligences that propose and the systems that execute
For every proposed action, it:
- 01
Checks the explicit constraints
Safety, physics, operations, regulation. Written, versioned rules that can be held against you — not an implicit judgement buried inside a model.
- 02
Authorises, conditions or stops the action
Graded verdicts: the action goes through, it goes through under conditions, it is referred to a human, or it does not happen.
- 03
Sets an autonomy level decision by decision
Autonomy is not a global setting fixed once and for all: it is assessed decision by decision, according to the risk and the system’s actual track record.
- 04
Seals a structured, reproducible and auditable record
What was decided, on what data, under which constraints, with which alternatives set aside. Replayable identically, verifiable by a third party.
What DECIVER is not
DECIVER replaces neither the robot controller, nor the safety PLC, nor your production system. It does not certify anything in law. It supervises decisions — that is all, and that is what is missing.
The analogy
The air traffic controller
They fly no aircraft; they authorise, constrain, interrupt and document. Air traffic is not safe because every pilot is perfect: it is safe because an independent layer of authority rules on every movement and keeps the record of it.
03 Judge us on the evidence
Four verifiable facts, not four promises
01
Working software demonstrator
100% of actions go through a sealed and traceable decision.
02
Replayable identically
Same situation, same decision, same record — verifiable by a third party.
03
Generic by design
A second industrial domain integrated without altering the decision core.
04
Documented limits
We publish what the system cannot do yet.
We would rather demonstrate than promise.
Demonstration on request04 First demonstrator
Adaptive robotic battery disassembly
First field of application: adaptive robotic battery disassembly — a demanding environment where electrical safety, physical integrity, the circular economy and regulatory traceability all meet.
Every move there is irreversible, every pack differs from the last, and a mistake cannot be undone. That is precisely why we started there.
The same decision layer applies to
Robotics integration
Multi-robot cells, human-machine co-activity, frequent changeovers.
Logistics and intralogistics
Autonomous fleets, materials handling, shared and shifting environments.
Drones and autonomous systems
Inspection, intervention, missions under strict regulatory constraints.
Maintenance in demanding environments
Energy, continuous processes, sites where unplanned downtime costs the most.
The decision core is independent of the domain. What changes is the constraints, not the layer that enforces them.
05 The category
Physical Decision Intelligence
After generative AI and embodied AI, the next step is not better models — it is better decisions.
PDI is the decision intelligence layer for complex physical systems: it does not replace your technologies, it governs them.
06 Who we are
Ahvelior Partners
Ahvelior Partners is a French SASU based in Lyon. We build DECIVER and the Physical Decision Intelligence layer underneath it.
Alberto González Pizarro, president, leads the product work and the conversations with industrial partners.
07 Contact
Let’s talk about your decision chain
If your machines already carry out actions proposed by an AI system — or if you are looking at the subject for the next two years — write to us. We reply directly, with no one in between.
Technical demonstration on request, by video call or on site.