Synchronized with reality

Team up with AI agents. Stay in charge.

Metronome keeps one shared picture of your operation: orders, capacity, constraints, commitments, current as things change. Your people and your AI agents both work from it, so their decisions fit together instead of colliding.

Get started Explore the platform
Process Metronome as the operational graph Agents sit above a governed layer made of the operational graph, bounded dispatch and a single audit trail. That layer reads from the systems of record below it: ERP, WMS or TMS, email and files, and spreadsheets. AGENTS Planning agent Dispatch agent Monitoring agent PROCESS METRONOME Operational graph Facts, constraints, dependencies Bounded dispatch One action, one named owner Single audit trail AI and human, same record YOUR SYSTEMS ERP WMS / TMS Email & files Spreadsheets Nothing is replaced. The layer reads what you already run.

The problem

The plan in one file. The ground truth in another.

Operations rarely fail because someone made a bad call. They fail because the plan lives in one place and what's actually happening lives in another. By the time the two are reconciled, the decision has already been made on the wrong information.

The plan goes stale

A plan is a snapshot. The operation keeps moving after it's taken, and nobody updates the file until the next planning cycle. By then it's already wrong.

The facts are scattered

The commitment sits in one system, the capacity in another, the constraint in a spreadsheet nobody else has open. No one view holds all three at once.

AI doesn't fix this. It speeds it up.

An agent working from a stale or partial picture doesn't notice. It acts on what it's given, quickly and confidently. The agent isn't the problem. What it's reading is.

Metronome keeps one picture of the operation in step with what's actually happening, and both your team and your AI agents work from it: same facts, same constraints, same moment. That's what lets their decisions fit together.

Metronome agents plan01 against what's true right now, decide02 within limits you set, and act03 into a single record.

01

Plan against what's current

Your orders, commitments and capacity become one connected picture. Agents plan against that, not against last week's export.

02

Decide within limits you set

Something changes. The agent works out what it breaks downstream and brings one proposal to the person who owns that call.

03

Act with a full record

One action at a time, inside the limits you set. Approved or turned down, everything lands in the same record.

01

Agents plan against what's current

Describe what matters once: your orders, your commitments, your team's capacity. Metronome builds the connected picture from what you described, keeps it current, and every agent plans against that instead of a stale export. It doesn't write your plan for you, and it doesn't replace your ERP, WMS or TMS. It reads from them, and makes sure the facts your plan rests on are still true.

Planning coverage

  • Track orders from intake to delivery
  • Match work to the capacity you have
  • Plan recurring demand, and spot it drifting
  • Hold the constraints your plan has to respect
  • See what depends on what, before it breaks
  • Carry commitments that repeat
The operational graph Orders, vessels, plants, lock schedules and crew rotas modelled as connected nodes, with the constraints that link them made explicit rather than held in someone's head. blocks arrival Order #4417 Due Thu 06:00 Barge Aurore Capacity 1,200t Lock Suresnes Closed 22:00-05:00 Plant 7 Stock 2.1 days Crew rota 2 of 3 available

02

Decisions route to the right owner

Send signals, not prompts: a new order, a status change, a deadline that moved. The agent reads the updated picture, works out what that changes downstream, and brings one proposal to the person who owns it. What an agent can do on its own and what needs your sign-off is yours to set, and the platform holds that line for you.

Decision routing

  • Catch conflicts as they appear
  • Flag overcommitment before you accept it
  • Show what a change breaks downstream
  • Route each proposal to whoever owns it
  • Set what needs sign-off and what doesn't
  • Attach the evidence to every proposal
Signal to proposal to owner An inbound signal is evaluated against the graph. The agent proposes one action with its evidence, and it is routed to the named owner, who confirms or dismisses. SIGNAL Lock Suresnes closes 22:00 Agent reads the graph: 11 deliveries downstream, 2 carry penalties PROPOSAL Resequence 11 deliveries, hold Aurore until 05:10 Protects both penalty commitments. Plant 7 stock holds to 2.6 days. Confirm Dismiss Owner: M. Report, duty planner Nothing moves until a named human decides. A dismissal is recorded too.

03

Actions are dispatched, audited, yours

The agent acts one step at a time, inside the limits already set, and the rest happens quietly: watching, recalculating, flagging, waiting. No chat window asking for your attention. Every action, human or AI, lands in the same record. You can widen what agents handle on their own as trust builds. It isn't a switch you throw on day one.

Execution and record

  • Dispatch one step at a time
  • Watch for drift after you approve
  • Recalculate the moment something shifts
  • Keep one record for people and agents alike
  • Trace every action to a named person
  • Move nothing until an owner confirms
One audit trail A single chronological record in which agent actions and human decisions are interleaved, including dismissals, with no separate log to reconcile. AUDIT TRAIL 05:12 AI Planning agent Recalculated 11 delivery windows 05:12 AI Planning agent Flagged 2 penalty commitments at risk 05:19 YOU M. Report Confirmed resequence proposal 05:19 AI Dispatch agent Dispatched revised rota to 3 crews 07:40 YOU C. Duval Dismissed reroute: barge already loaded 07:41 AI Monitoring agent Held plan, watching Plant 7 stock One record. No separate AI log to reconcile at audit time.

Why this is adoptable

The agent resolves it in milliseconds. Your team feels the fifth replan.

AI agents and people don't experience time the same way. An agent settles a conflict in milliseconds and feels nothing. A team that has already absorbed four replans today feels the fifth. Metronome paces what the AI asks of your team to what the team can actually take: one proposal at a time, inside the workflow they already follow, owned by someone you can name.

There's no chat window and no prompts to write. The AI works inside the actual task, and stays quiet until it has something worth your attention.

Running in live pilot operations

An industrial minerals supply chain on the Seine.

Tugboats, barges and twelve concrete plants, running against lock schedules. An idle barge costs on the order of €3,000 a day; a plant without material costs far more. Metronome keeps all of it in one picture that stays current: the AI proposes, the planner approves with one click, and every decision can be traced back to the evidence behind it.

Read the full case study

Where to start

Same system. Different stakes.

Replan without rebuilding the plan

Replan in minutes, not in a spreadsheet.

COO, Operations Engineer, Head of Planning

See Operations

Trust and control

Adopt AI without losing the audit trail.

CIO, Head of AI, Risk & Compliance

See Governance

Connect your agents

Give your agents structure, safely.

CTO, AI Engineer, Platform Lead

See Engineering

Start where you are.

Describe the part of your operation you run today and put it to work this week. The same thing that keeps one person's week straight keeps a logistics network straight. What changes is how much you describe.

Get started

Metronome is in beta. You create a workspace and start with your own work - nothing to install, no data migration to begin.