For enterprise
An operation that cannot stop, running with AI agents inside it.
Metronome is in live pilot operations on the Seine: barges, twelve concrete plants and the lock schedules they all depend on, replanned with AI proposals that a named planner approves. This page is for the conversation that comes before a deployment - what the platform enforces, what it records, and what it does not do yet.
Sixty minutes with the founders. We map a piece of your operation into the graph and you leave with a concrete architecture proposal.
Running in live pilot operations
An industrial minerals supply chain on the Seine.
Tugboats, barges and twelve concrete plants, running against lock schedules and storage levels that set delivery urgency. An idle barge costs on the order of €3,000 a day; a plant without material costs far more. Replanning that took a morning of calls and spreadsheets now takes minutes, with the reasoning attached, and the planner still approves every move.
The capacity and on-time targets we are working towards in that pilot are stated on the case study page, labelled as what they are: objectives, not yet a published result. We will publish them when they are validated and the client has signed off on the language.
Read the full case studyWhat the platform enforces
Seven constraints, enforced by the platform rather than by a policy someone can bypass.
An agent cannot express an action outside its mandate, which is what lets you show an auditor that it could not have exceeded its bounds, not merely that it did not. The limit worth stating in the same breath: the agent can still reason imperfectly inside its frame. What it cannot do is act outside it.
The agent knows what to do, because you set it.
What an agent is aiming at comes from the work in front of it. The platform sets the objective; the AI does the reasoning.
Every action is expected. No surprises.
The step came from a process you defined. Teams accept change they saw coming. They reject change that appears from nowhere.
The agent cannot make an invalid move.
An action that would break your rules was never on the list the agent could choose from.
Decisions stay close to the ground.
Work routes to the person closest to it. The AI supports local judgment. It does not overrule it.
One action at a time. Always.
One interaction, one outcome. No chains of autonomous decisions. After each action, the system looks again.
AI and people share one record.
No separate AI log to reconcile later. Both land in the same steps, with the same timestamps.
A person always owns the outcome.
Every AI action traces back to someone you can name. Accountability never goes missing.
Compliance posture
What exists today.
Available on the Enterprise tier
- Single sign-on
- On-premise deployment
- Full audit-trail export
- Role-based views and access
- Custom graph constraints
- Data residency controls
What we do not claim
We hold no external certification of the audit trail, and we will not imply one. We do not make language models truth-preserving; nothing does, which is why the bounds sit in the platform instead of in the model. We do not replace your ERP, WMS or TMS - the graph reads from them. And we do not claim AI adoption is a solved problem. We claim the structural preconditions for it, and we are proving the rest in live operations, one vertical at a time.
Material
To read before the conversation.
The case study covers the pilot in full. The one-pagers for operations and governance, and the white paper on bounded AI in operations, are being written - ask for them in a working session and we will send what exists today rather than a placeholder.
What this is not
- Can we run this on our own infrastructure?
- On-premise deployment is part of the Enterprise tier. Talk to us about your constraints before you plan around it, so we can tell you what is available today rather than what is on a roadmap.
- Has the audit trail been through an external audit?
- Not yet. What exists today is the record itself - human and AI decisions, approvals and dismissals, in one chronological trail - and a live pilot exercising it daily. We would rather tell you that than imply a certification we do not hold.
- Do we have to replace our ERP, WMS or TMS?
- No. The operational graph reads from the systems you already run. Nothing is replaced, and nothing needs to be migrated before you can see it work.
- What does a working session actually involve?
- Sixty minutes with the founders. We map a piece of your operation into the graph and you leave with a concrete architecture proposal. It is not a product demonstration, and there is no deck.
Bring us an operation that cannot stop.
Sixty minutes with the founders, on your operational context rather than our product. You leave with a concrete architecture proposal; we leave with a clear picture of fit.