Skip to content
Contact

Service · AI automation · Agents

AI automation: For processes that cost hours today.

Answering enquiries, entering orders, reading datasheets, writing test reports: many processes are easy to describe and still cost hours every week. elsyQ automates them with agents that operate your systems. Where a rule exists, the rule decides. The language model reads what no rule captures.

Describe a process How a case runs

At a glance

Typical processes
Enquiries, orders, datasheets, reports, test sequences
Integration
ERP · inbox · DMS · PLM · instruments
Control
Permissions, approvals and a complete log
Models
Local on your hardware or through OpenAI, Anthropic and Mistral

From the starting point to the result

  1. Starting point

    • Staff copy data by hand from one system into the next.
    • Classic process automation fails on e-mails, PDFs and free text, because every input looks different.
    • Errors surface late. How long a case takes depends on who happens to have time.

  2. Solution

    • A language model reads the input and transfers it into a fixed schema.
    • Rules, master data and limits check the result, the same way every time.
    • Whatever fails the check goes to a person, together with the reason.

  3. Integration

    • Into your systems through APIs, file shares or the inbox.
    • In engineering also directly to plant and instrument: OPC UA, MQTT, SCPI.
    • Runs on your infrastructure. The model runs locally or comes through an API.

  4. Result

    • Your team only handles the exceptions.
    • Every case is logged and can be traced back.
    • A before-and-after comparison shows how much time the process really saves.

What we automate

Many of the examples come from engineering: electronics, machinery and measurement. The processes behind them exist in every company: something comes in, someone reads it, checks it and enters it somewhere.

Enquiries and quotes

Read enquiries from the inbox, match them against price lists and master data and create a draft quote that your sales team approves.

Orders

Recognise orders sent as PDF or e-mail, check them against the item master and terms, and create them as sales orders in the ERP.

Datasheets and technical documents

Extract parameters from datasheets, standards and test reports, with unit and source, and transfer them into your database.

Reports

Merge data from several systems into a report in which every value points back to its source.

Test and measurement sequences

Drive instruments through a test matrix by script, check the results against limits and generate the test record.

Monitoring measurement and machine data

Detect anomalies in measurement series and machine data and report them so that the service team understands what happened.

How a case runs

An order arrives by e-mail. What happens next.

The model reads, the rules decide. That split runs through every automation we build, because it is the only way to say afterwards why a case went the way it did.

  • Language model: reads and proposes
  • Fixed rules: check and decide
  • Person: exceptions only
  1. Input

    A PDF in the inbox

    An order arrives as a PDF attachment. Layout, language and item names differ from customer to customer.

  2. Read

    Recognise the line items

    The language model reads items, quantities and delivery dates and enters them into a fixed schema. That is all it may do at this point.

  3. Check

    Match against your data

    Item numbers against the item master, prices against the terms, dates against capacity. Fixed rules that always decide the same way for the same input.

  4. Exception

    To your team, with the reason

    If something does not match or the model is unsure, the case goes to a person, together with the reason. Everything else runs through.

  5. Book

    Create the order, log the step

    The order is created in the ERP. Input, model output, check result and decision stay on record.

Why we build it this way: determinism and AI

AI agents within a fixed frame

An agent only works with the tools you give it.

An AI agent is a language model that does not just answer but acts: it reads an e-mail, queries the ERP and creates a case. To keep that under control, we build a frame around every agent, the agent harness. It defines which tools the agent may use, which data it may see and when it has to wait for an approval.

Every step is logged. Before an agent works in production it runs against test cases from your daily work, and only once it passes them does it take on real cases. The language model behind it stays replaceable.

From our projects

The same principle is at work in projects we have built: the model helps to read and explain, the numbers come from measurement and computation.

Client project · RF measurement

VNA software with an AI assistant

An AI assistant answers questions about measurements and component models in plain language and points out anything unusual. The measured values themselves come from the instrument.

View the project: VNA software with an AI assistant

Client project · IoT

Field-device data platform

Readings arrive over MQTT, a Laravel application evaluates them in real time and shows the service team which device needs attention. In the client’s own data centre.

View the project: Field-device data platform

Own product · In development

Circuit Forge

An assistant proposes circuits and values, and a deterministic engine checks whether they hold. In the default setting, nothing is applied until a person agrees.

View the project: Circuit Forge

Process

Start small, measure, then expand.

We automate a single process first and measure beforehand what it costs today. You decide after every phase whether we continue.

The process in detail
  1. → brief

    Analysis

    We look at the process where it happens and record how much time it costs today.

  2. → running prototype

    Prototype

    A first version runs on your real data, on request with a local model on your hardware.

  3. → your sign-off

    Presentation

    We show the prototype to your team and set time saved and error rate against the baseline.

  4. → live operation

    Production

    The solution goes live, built into your systems, with permissions, logging and monitoring.

Common questions

How is AI automation different from classic process automation?

elsyQ uses AI where classic process automation such as RPA reaches its limit: with input that looks different every time, such as e-mails, PDFs or free text. RPA replays fixed click sequences and stops as soon as something deviates. In our systems a language model reads the input, and fixed rules check the result before it reaches your system.

What is an AI agent?

At elsyQ an AI agent is a language model that calls tools instead of only answering: it reads an e-mail, queries the ERP and creates a case. We build agents inside a fixed frame, the agent harness, which sets permissions, approvals, limits and logging. That keeps it traceable what the agent did and why.

Which processes are suited to automation?

elsyQ automates processes that are easy to describe, happen often and still cost time today, for example enquiries, orders, reading datasheets, reports or test sequences. Decisions for which there are neither rules nor examples are a poor fit. Whether a process is worth it is settled in the analysis, before any quote is written.

How safe are automatic decisions?

At elsyQ no language model decides anything that money, safety or a measurement depends on. The model reads and proposes. Checking and deciding is done by fixed rules, and uncertain cases go to a person together with the reason. Every step is logged.

Does AI automation work without the cloud?

Yes. elsyQ runs the language models entirely on your hardware if you wish, so no data leaves the building. Where that is not necessary, we connect OpenAI, Anthropic or Mistral through an API. The architecture keeps the switch open, so you are not tied to one provider.

Does elsyQ only work for engineering companies?

No. elsyQ works for electronics and semiconductor companies, machinery and plant engineering and measurement technology, but automates processes for companies in any industry. What matters is not the industry but whether the process can be clearly described.

Let’s talk

Which process costs your team hours every week?

Describe it in a few sentences: what comes in, what happens to it and which system it ends up in. We will tell you whether and how it can be automated.

Describe a process View our work