Skip to main content

AI agents that work in your factory.

Frontlink AI works with tools on your live production data. It looks things up, checks the numbers and prepares changes. You confirm, Frontlink carries them out and records everything.

An agent at work.

The production manager is looking at Filling line B and asks one question. Follow what the agent does: which tools it picks up, what it finds and which change it prepares.

  1. 1 / 5You ask from your screen

    The agent knows which line you are looking at and fetches that data again itself.

  2. 2 / 5It picks up its tools

    Live status, losses, orders at risk. You see each step while it runs.

  3. 3 / 5An answer from your records

    Frontlink calculates the numbers, not the language model. Every order and line in the answer links to the record.

  4. 4 / 5It prepares a change

    A proposal with exactly what changes, and what the effect is. Nothing has happened yet.

  5. 5 / 5You confirm, Frontlink executes

    Executed with your permissions, recorded with who asked and who confirmed.

You ask from your screen

The agent knows which line you are looking at and fetches that data again itself.

Example with fictional data.

The agent in planning.

The customer calls: order 88240 needs to be ready sooner. The planner asks the agent, right on the planning board. The agent finds the room, shows the proposal on the board and waits for approval.

  1. 1 / 4A question in plain language

    The planner asks from the planning board. The agent knows which week and which lines it is about.

  2. 2 / 4The agent works it out

    It fetches the status of 88240 and sees Filling line B running late because of 88213. Filling line A has room on Tuesday afternoon.

  3. 3 / 4The proposal is on the board

    Dashed: 88240 to Filling line A, Tuesday 14:00. The proposal also names the effect: 88236 moves up one slot.

  4. 4 / 4Confirmed, the board is updated

    The planner confirms. Frontlink moves the order with the planner's permissions and records who did what.

1. A question in plain language The planner asks from the planning board. The agent knows which week and which lines it is about.

Example with fictional data.

The agent in quality.

An afternoon of the quality manager, in four questions. From open checks to a batch record ready for release. The signature stays with a person.

  1. 1 / 4What needs attention

    This week's open, overdue and failed checks, and the scrap. From the records, not from a summary.

  2. 2 / 4Prepare the record

    The agent gathers lots, checks and deviations into a draft and sees what is still missing.

  3. 3 / 4Compare versions

    What changed since the previous version, by whom and when. Useful for the review.

  4. 4 / 4Release stays human work

    The agent proposes to finalize version 2. The quality manager confirms, with their own permissions.

What needs attention. This week's open, overdue and failed checks, and the scrap. From the records, not from a summary.

Example with fictional data.

Your own AI, the same tools.

Already working with Claude, ChatGPT or your own agent? It gets the same access as Frontlink AI: the same tools, the same permissions, the same approval. Your systems use the same route.

  • APIFor your systems

    A documented API your ERP, webshop or own scripts use to read and write.

    ERP · webshop · scripts
  • CLIFor your IT and agents

    Frontlink from the terminal. For your IT, and for AI agents that work in a terminal.

    IT · coding agents
  • MCPFor your AI assistant

    Connect Claude or ChatGPT to Frontlink. Ask about orders, planning and stock, and have changes prepared for your approval.

    Claude · ChatGPT · your own agent
The same rules for every AI
  • The same tools as Frontlink AI
  • Never more permissions than the user behind it
  • Changes wait for approval
  • Every action recorded, human or AI

An AI may never do more than the person behind it.

Improving your processes together.

The agent looks across all your production data: orders, planning, the floor, quality and the warehouse. That lets it find where time and material leak away. Your Frontlink engineer turns that into improvements with you.

  1. 1Measure

    Records from the floor, the machines and the warehouse come together, with how fresh and reliable they are.

  2. 2Search

    The agent connects the dots across areas: which machine, which product, which shift, which supplier.

  3. 3Improve

    Your engineer and your team choose the approach and change the process in Frontlink: a check, a step, a rule.

  4. 4Measure again

    Does it work? The trend shows it, and how certain that conclusion is.

Questions the agent can work out
  • Where does the time go on Filling line B?
  • Which machine produced the most scrap this month?
  • Which orders still lack evidence for the batch record?
  • Did OEE really go up, or is it the data?
And Frontlink gets better
  • Every question sharpens the tools

    Can the agent not answer a question well yet? Then we build the tool it needs, for you and for your own AI.

  • Tested first, then with you

    We test new tools on fictional data before they run in your environment.

  • An improvement analyst, built with you

    An agent that goes looking for waste by itself and comes back with proposals. We are building it together with our first customers.

Safe by design.

An agent that works inside your production has to meet stricter rules than a chatbot. These are the rules Frontlink AI works under.

Action request#1042
  1. Proposed
  2. Confirmed
  3. Executed
Action
Move order · 88240
Change
Filling line A, Tue 14:00
Requested by
Planner · 10:41
Confirmed by
Same planner · 10:42
Executed as
The planner, with their permissions
Conversation
Kept with the request
What the AI may do
  • Never more than the user

    Every tool checks the user's permissions, when proposing and again when confirming.

  • Nothing without approval

    Every change is a proposal. Only the person who asked can confirm it, and it then runs through the same route as a manual action.

  • Only approved data

    The agent reads from approved tables, in limited amounts. Sensitive fields are removed before anything is shown.

What the AI says
  • Numbers from your records

    Frontlink calculates counts, times and stock. The language model repeats them and does not do its own maths.

  • Honest about the data

    If data is stale or incomplete, the agent says so.

  • Tricks are refused

    The agent refuses requests to ignore its instructions, bypass permissions or hand over all data.

What is recorded
  • Every request recorded

    Who asked, who confirmed, when, what changed and whether it worked. The conversation is kept with it.

  • Separated per company

    Your data is isolated per company in the database. The agent can only reach yours.

  • You choose the model

    Your administrator chooses which AI model your environment uses and can adjust the agent's instructions.

One assistant, in every part.

Frontlink AI is not a separate module. It works on the same data as orders, planning, production, quality, warehouse and performance, so it understands how they connect.

  • Quotes and costingCalculates a quote from your recipes and prices, and finds products quoted below their real cost.
  • Orders and customer contactLooks up orders, shows progress and expected finish, and moves an order to the right kanban stage.
  • Customer portalShows which customer is waiting for what and drafts the update or reminder for your approval.
  • PlanningShows which orders are at risk and proposes to move an order, bring it forward or send it back.
  • Production and operatorsShows what is running on every line now and adds downtime reasons.
  • Quality and batch recordShows open and failed checks, prepares the batch record and compares versions.
  • Warehouse and stockLooks up items, lots and records and turns supplier documents into data.
  • PerformanceInvestigates losses per machine and shows the OEE trend, with how reliable the data is.

Frequently asked questions.

Can the AI change things by itself?

No. Every change is a proposal you confirm. Only then does Frontlink carry it out, with your permissions, and record who asked and who confirmed.

Does the AI make up numbers?

No. Numbers come from your records and are calculated by Frontlink. The agent repeats them, links to the records and says so when data is stale or incomplete.

Which AI models does Frontlink use?

By default a model from Anthropic. Your administrator can choose another model per environment from a fixed list, including OpenAI models.

Where is our data processed?

The agent only sends the data a question needs to the chosen model. Which provider and which region we use is agreed per customer.

Can we connect our own AI?

Yes, through the same route as Frontlink AI. The API is available for the warehouse and we extend it area by area. Claude and ChatGPT connect through MCP. Your own AI never gets more permissions than the user behind it.

Do we need to set anything up?

We set up Frontlink with you around your processes. The AI then works on the same data straight away, without a separate project. If a tool is missing for your process, we build it with you.

Which process takes you the most time?

Book a 30-minute call. We show what Frontlink does with your own examples.

Book a call