Agent Platform for manufacturing

Fix the machine before it fails, without your data leaving the plant.

Agents read your downtime and sensor data, plan the repair around production and wait for a person before they book anything. The platform runs in your environment, so IT and OT keep control of data, access and cost. See it in a 30-minute walkthrough.

Trusted by manufacturers

  • adidas
  • EGGER
  • WEPA
  1. SignalSensor data

    The drive bearing on press PR-07 runs at 84 °C, up from 61 °C in one week.

  2. FindProduction Analyst

    PR-07 lost 405 minutes in seven days, 290 of them unplanned. The bearing is the cause.

  3. PlanMaintenance Planner

    Repair on Saturday from 08:00 to 12:00. Parts are in stock and the crew is free.

  4. WaitWork Order Assistant

    Create work order WO-48213 in the maintenance system? It waits until a person allows it.

  5. DoneMaintenance system

    WO-48213 is released. The bearing kit is reserved and two technicians are booked.

01 / Why

Your plant already has the data to see the next failure coming.

It sits in the MES and the historian. The people who can read it are retiring or busy. The data itself must not leave the plant.

  1. Unplanned stops.

    Repairs planned around production.

    Agents check downtime and sensor data on every shift and find a window before the machine fails.

  2. Know-how retires with your best technicians.

    Their know-how stays in the plant.

    An engineer describes in plain words how to diagnose a fault. The agent applies it on every shift.

  3. Not enough people in IT and OT.

    Plant teams build their own agents.

    From connectors IT approved once. A forward deployed engineer builds the first agents with them.

  4. Plant data must not leave the site.

    The platform runs in your plant.

    On your hardware or in your cloud, with an open-weight model if you want. This also works at sites without an internet connection.

  5. Locked into one cloud's AI.

    Yours to keep.

    Open source and built in the EU. It runs on the Kubernetes you already have and stays with you if the contract ends.

Built once on line 3, reused in every plant.

  1. Line 3

    A forward deployed engineer sets up the platform in your environment and builds the first agents with your team.

  2. The whole plant

    The next line reuses the agents and connectors of line 3. The planner adapts the instructions to its machines.

  3. Every plant

    The second plant starts from what the first one built. There is no second setup.

02 / How

One press at risk, repaired before it fails.

The maintenance team built the three agents in the demo. Each one shows its work and waits for a person before anything changes. All data in the demo is fictional.

  1. Signal

    Sensor data

    The drive bearing on press PR-07 runs hot.

    Its temperature climbs from 61 to 84 °C within one week.

  2. 01 Find

    Production Analyst

    The machine that costs the most time

    The analyst reads the downtime report and the sensor trend. Press PR-07 lost 405 minutes in seven days, 290 of them unplanned. The worn drive bearing is the cause.

    • Read-only
    • Every query logged
  3. 02 Plan

    Maintenance Planner

    A repair window that costs no customer order

    The planner checks the production schedule, the spare parts store and the crew. It proposes Saturday from 08:00 to 12:00, right after the die change.

    • Parts in stock
    • Crew available
  4. 03 Wait

    Work Order Assistant

    The work order waits for one Allow.

    The assistant shows the exact order it wants to create in the maintenance system and pauses. It books nothing until a person clicks Allow.

  5. 04 Done

    Maintenance system

    The maintenance system releases work order WO-48213.

    The order reserves the bearing kit and the shaft seal and books two technicians for Saturday morning. The approval and the order are in the audit trail.

03 / What

Agents on the systems your plant already runs, under rules IT sets once.

Connectors use the open Model Context Protocol. Teams reuse existing connectors. A forward deployed engineer builds the missing ones with your team.

How the agents reach the plant systems The Agent Platform runs in your environment with three agents: Production Analyst, Maintenance Planner and Work Order Assistant. They read from the MES, the historian and the shift plan. They write to the ERP, for example SAP PM, and to the maintenance system only after a person allows it. In your environment Reads MES Orders, line status and downtime reasons. Reads Historian Sensor trends such as bearing temperature. Reads Shift plan Who is on shift with which qualification. Writes after approval ERP, for example SAP PM Spare parts stock and reservations. Writes after approval Maintenance system Work orders and the maintenance history. Agent Platform One set of connectors Production Analyst Maintenance Planner Work Order Assistant
  • MES

    Orders, line status and downtime reasons.

    Reads
  • Historian

    Sensor trends such as bearing temperature.

    Reads
  • Shift plan

    Who is on shift with which qualification.

    Reads

In your environment

Agent Platform

The three agents from the demo share one set of connectors.

  • Production Analyst
  • Maintenance Planner
  • Work Order Assistant
  • ERP, for example SAP PM

    Spare parts stock and reservations.

    Writes after approval
  • Maintenance system

    Work orders and the maintenance history.

    Writes after approval
  • Read access
  • Writes only after a person allows them

For IT and OT

Every agent a plant team builds inherits these rules.

Nobody configures them again for the next line.

  1. The platform and the models run in your environment.

    With an open-weight model on your own hardware, no plant data leaves the site. Sites without an internet connection get releases through an offline channel.

  2. Every write waits for a person.

    The agent shows the exact call first. The approval and the call land in the audit trail.

  3. Agents act with the permissions of the person who started them.

    An agent never gets more access than the engineer who uses it.

  4. IT approves every connector.

    Plant teams pick from the connectors IT released.

  5. Cost is visible per line, team and model.

    Each team has a budget and IT sees every model call.

04 / Proof

Manufacturers run their platforms with Giant Swarm.

The Agent Platform comes from the team that has run production platforms for manufacturers for more than twelve years.

  • 12+ yearsrunning enterprise platforms in production
  • ISO 27001certified information security management
  • Apache 2.0components in the CNCF and the Linux Foundation
Giant Swarm allows our small team to offer a world-class platform for all our modern applications and AI workloads.
Simon Gellrich Manager Web Development, EGGER

05 / Questions

Frequently Asked Questions

How do factories safely introduce AI agents into existing production systems?

Factories introduce AI agents safely by letting them read plant data first and change nothing without a person. On the Giant Swarm Agent Platform, agents read the MES, the historian and the shift plan, act with the permissions of the person who started them, and write to the ERP or the maintenance system only after someone allows it. The Giant Swarm guide to AI agents in manufacturing goes deeper.

Can AI agents run without plant data leaving the site?

Yes. The Giant Swarm Agent Platform runs in the plant's own environment, on its own hardware or in its own cloud. With an open-weight model on the plant's hardware, no plant data leaves the site. Sites without an internet connection get platform releases through an offline channel, so the agents also work where there is no outside connection.

Which plant systems can the Agent Platform connect to?

The Agent Platform connects to the systems a plant already runs, through connectors based on the open Model Context Protocol. Agents read from the MES, the historian and the shift plan, and write to the ERP, for example SAP PM, and to the maintenance system after approval. IT approves every connector once, and a forward deployed engineer builds missing ones with the plant team.

How do AI agents help with predictive maintenance?

AI agents check downtime and sensor data on every shift and find a repair window before a machine fails. On the Agent Platform, one agent finds the machine that loses the most time, a second plans the repair around production, spare parts and crew, and a third prepares the work order, which waits until a person allows it.

How do manufacturers scale AI agents from one line to every plant?

Manufacturers scale AI agents by building them once and reusing them. A Giant Swarm forward deployed engineer sets up the Agent Platform and builds the first agents with the team on one line. The next line reuses those agents and connectors, and the second plant starts from what the first one built, without a second setup. The first line starts with the Foundation Package at a fixed price.

See it on one of your lines.

A 30-minute walkthrough on a lab installation with one of your maintenance or planning workflows, including the approval and the bill at the end. The first line starts with the Foundation Package at a fixed price.