Industrial AI for manufacturing

    The complete AI system for manufacturing companies to digitalise and automate your business processes.

    Data, models and processes in one system, on your own hardware. Efficient on three levels, with one goal: the self-learning machine.

    Our thesis

    Making data and AI inference for industry as efficient as possible.

    That is the goal. Efficiency on three levels that only make the difference together.

    01

    Data efficiency

    One continuous data pipeline instead of five siloed tools. From raw signal to dashboard, from a single source.

    How we do it

    Machine, MES and ERP data as well as documents are captured at the source, validated and turned into reliable decision data through release stages.

    02

    Inference efficiency

    Small, domain-specific fine-tuned models instead of large general-purpose LLMs. More usable answer quality per watt and per euro of hardware.

    How we do it

    Like a drive sized for its load: our own OPAIRS models for SQL, production, maintenance and time series, quantised and trained on HPC. Measured like a plant, with quality, response time and availability per use case.

    03

    Cost efficiency

    One-off investment instead of usage-based fees. Cost per user goes down as usage spreads across the company, instead of going up.

    How we do it

    Hardware and software as one system, with no token costs and no subscription. SLA and maintenance are flexibly tailored to your operations, and the hardware grows with more GPUs when needed.

    Sovereignty is the prerequisite. Efficiency is the business.

    The customer keeps the data and development advantage and digitalises the business process with the highest value creation.

    Cyber-physical AI · How we understand AI

    AI as a cyber-physical system

    We do not see AI as a detached agent but as a cyber-physical system. It captures data from machines and processes, evaluates it with specialised models and is designed to interact with its environment under control: with clear limits, approvals and traceable decisions.

    1. 1

      Capture

      Machine, process and business data in one pipeline.

    2. 2

      Evaluate

      Specialised models with measurable quality per use case.

    3. 3

      Act under controlIn development

      Our goal: an ecosystem that gives agentic AI in production a controlled environment, with guardrails and approvals built in.

    More on cyber-physical AI →
    OPAIRS AI system architecture: cyber-physical, local industrial AI

    Where we are heading

    Towards the self-learning machine

    A machine that learns from its own operation: that is the goal we pursue together with you. Approvals and outcomes are already captured today, and models are tuned specifically to your use cases. Step by step, this leads to workflows that run reliably and automatically. The compute for this sits right at your site.

    1. 1

      Understand

      Bring together data from plant, ERP and documents.

    2. 2

      Assist

      Answers, forecasts and recommendations your team approves.

    3. 3

      Learn by itselfGoal, together with the customer

      Together with you, we evolve the system so that it learns from approvals and outcomes and automates workflows.

    Example imageExample image: HP Z8 Fury G5 workstation with multiple NVIDIA GPUs
    Example image: HP Z8 Fury G5 with multiple NVIDIA GPUs. Configuration depends on the use case.

    Hardware platforms in use

    With NVIDIA GPUs, from compact workstations to servers. Not tied to a single vendor.

    Explore OPAIRS hardware

    Trusted by partners and customers

    Latest articles

    Insights

    An overview of industrial AI, data sovereignty and regulatory requirements that take AI seriously.

    Diagram: Excel plan, SAP, PLM, Microsoft Teams and tracking tools flow within the internal company network into the OPAIRS data lake with knowledge hub, BI dashboard and AI agentUse Case

    From Excel Plan to Critical Path: How a Special Machinery Builder Manages Its Projects On-Premise

    The project plan lives in Excel, while the actual status is spread across SAP, PLM, Teams and the lists kept by each department. A special machinery builder with around 100 million euros in revenue brings both together with OPAIRS: plan and actuals are compared, the critical path becomes visible and the AI derives recommendations. The result: up to 10% fewer project hours and up to 15% less search time. All data stays in-house.

    Read article
    Stylised CNC milling machine, injection moulding machine and control cabinet with a sensor, connected by cables labelled OPC UA, S7, ADS, Modbus and IO-Link to a data server, with a monitor showing a dashboard and a speech bubble on the right – connecting cyber-physical systems on-premiseArchitecture

    Cyber-Physical Systems with OPAIRS®: Connecting Machines and Sensors On-Premise

    So far, OPAIRS has learned from ERP systems, documents and databases. With OPAIRS Connect Machine, the machines themselves join in: controllers, sensors and equipment deliver their values via OPC UA, Siemens S7, Beckhoff TwinCAT, Modbus, EtherNet/IP or IO-Link into a local database that the OPAIRS® Dashboard and AI agents read from. Everything runs in-house, without the cloud and without a connection to external services.

    Read article
    Stylised server rack, stacked containers, a shield with a check mark and a bill of materials under a magnifying glass in turquoise – software bill of materials and vulnerability monitoring on-premiseRegulation

    OPAIRS SBOM Scanner: Continuous Vulnerability and Supply-Chain Monitoring for the Cyber Resilience Act

    Since 11 September 2026, manufacturers must report actively exploited vulnerabilities in their products within 24 hours – from December 2027 the Cyber Resilience Act applies in full, including a software bill of materials. To report, you first need to know what is inside your own software. OPAIRS built a scanner for that: it runs entirely on-premise, watches continuously and only raises the alarm when it matters.

    Read article
    Stylized factory with chimneys and tanks in turquoise, next to a bar chart with a rising forecast curve and gears – time-series forecasting for industrial manufacturingResearch & Development

    OPAIRS Forecasting: Self-Hosted Time-Series Forecasting for Manufacturing

    How much material will we consume next quarter, when will energy demand rise, which items will run out? Time-series forecasts answer questions like these – usually through cloud services that production data has to be sent to. OPAIRS has built its own forecasting service that runs entirely on-premise, recommends a model only when it demonstrably beats a simple baseline, and connects with the SQL Agent into one continuous chain: query the data, forecast, evaluate.

    Read article
    OPAIRS SQL Agent benchmark: pass rate of all six LoRA adapters for PostgreSQL, T-SQL and Apache Iceberg compared with Claude Opus 4.8, GPT-OSS-20B and the untuned base modelsResearch & Development

    OPAIRS SQL Agent: Comparing Six LoRA Adapters for Industrial Databases

    Six LoRA adapters, a 26-question catalog spanning PostgreSQL, T-SQL and Apache Iceberg databases, two external reference models: OPAIRS has systematically evaluated its SQL agent. Two Granite adapters lead the field. For production use, however, the deciding factors are not only answer quality but also speed, memory footprint, concurrency and the available context from ERP, MES, PLM and other industrial systems.

    Read article
    OPAIRS Runtime 3 on NVIDIA RTX PRO 4500 Blackwell with GPT-OSS-20B and up to 2,637 tokens per secondResearch & Development

    RTX PRO 4500 Blackwell: 3.4x LLM Throughput Through Runtime Optimization

    Same GPU, same main model, up to 3.4x the output: OPAIRS Runtime 3 raises the throughput of GPT-OSS-20B on the RTX PRO 4500 Blackwell to up to 2,637 tokens/s. At the same time, the tests show why Qwen3.8-27B will not take over the production stack for now.

    Read article
    NVIDIA Inception Program badge – OPAIRS Systems is a program memberPartnership

    OPAIRS Joins the NVIDIA Inception Program

    GPU-accelerated LLM inference on-premise: what NVIDIA Inception means for the OPAIRS stack, and why sovereign AI and high-performance hardware must go hand in hand.

    Read article
    OPAIRS EnergyManager in cooperation with aWATTar – intelligent electricity price scheduling for on-premise AI inferenceEfficiency & ROI

    When AI jobs can wait, they pay less for power. OPAIRS does it automatically.

    Electricity is not a fixed cost: it fluctuates hourly. OPAIRS uses the aWATTar API to shift AI inference and automations into low-price windows. The result: measurable returns, predictable costs, and no loss of functionality.

    Read article

    Integrations

    Connected to your existing system landscape

    From the mailbox to the ERP, from the models to the data platform: OPAIRS runs in your own infrastructure and connects to the systems you already use.

    Email

    OPAIRS reads unstructured information directly from your team's mailbox via Microsoft Graph and stays in the ongoing exchange with suppliers and customers: queries, commitments, rescheduled dates. Everything runs in your infrastructure, nothing leaves your premises.

    • Outlook / Microsoft 365
    • Gmail

    Unstructured data

    OPAIRS unlocks data from documents that are still processed by hand: order confirmations, delivery notes, invoices, bills of materials and packing lists, as PDF, Excel or photo. The content is embedded, made searchable and linked to your master data.

    • Word
    • Outlook
    • PDF
    • Excel

    ERP systems

    OPAIRS connects to your ERP, e.g. SAP, Infor or proALPHA, through their official interfaces (REST, OData, SQL). Your ERP remains the leading system; changes are only written back after approval.

    • SAP
    • infor
    • proALPHA
    • & more
    • OPAIRS Guardrails

    Models that run on your side

    Open language models (Mistral, Qwen, gpt-oss) and dedicated OPAIRS models run on your hardware: text-to-SQL for your data, production and maintenance, time-series forecasts and embeddings for search. You choose the model, your data stays with you.

    • Mistral
    • Qwen
    • gpt-oss
    • Embedding
    • OPAIRS SQL
    • OPAIRS Production
    • OPAIRS Maintenance
    • OPAIRS Time SeriesMCP

    Data platform

    Structured data in PostgreSQL and Apache Iceberg, vectors in Qdrant, files in SeaweedFS, connected through Apache Spark. An open, vendor-independent stack that you can operate and audit yourself.

    • PostgreSQL
    • Qdrant
    • Apache Iceberg
    • SeaweedFS
    • Apache SparkConnector

    All trademarks and logos mentioned belong to their respective owners. Their mention does not imply a partnership or certification.

    Our Products

    No SaaS. No subscription. An AI system you own, just like your machines.

    Entry / Pilot

    Phase 1: System setup & productive start

    Build a sovereign AI and data foundation in your own infrastructure.

    • All-in-one system (hardware & software)
    • Integration of existing data sources
    • AI only where it creates real value
    • Short time-to-value with clear scope
    • On-premise, no cloud dependency
    • Expandable and scalable
    Scaling

    Phase 2: BI & knowledge development

    Systematic expansion of analytics, knowledge and decision capability.

    • End-to-end data governance
    • Multiple use cases and departments
    • Verified AI and LLM models
    • BI, forecasting and process optimization
    • Compliance-ready (EU AI Act, ISO, NIS2)
    • Integration into existing IT/OT landscapes
    Customization

    Phase 3: The self-learning machine

    Developed together with you: models that get more precise with every use case, and workflows that run automatically under control. Aligned with strategy and culture.

    • Digital process twin
    • Tailored system architecture
    • Edge AI and hardware integration
    • Automation and robotics
    • Highest security requirements (incl. air-gapped)
    • Long-term extensible and future-proof

    Why OPAIRS?

    AI only where it creates real value

    Applied with your company knowledge and clear business goals.

    Trustworthy AI: data sovereignty & security by design

    Trustworthy AI, run locally: controllable, traceable and built for high security requirements.

    Cyber-physical, not detached agents

    AI that interacts with plant and process under control and makes operations transparent and steerable.

    Consulting, technology & execution

    From strategy to productive operation with one consistent partner.

    Single point of truth for your data

    Consistent data basis for analysis, automation and decisions.

    Ready to make your business and production processes as efficient as your machines?

    Consulting and analysis with clear technical and strategic recommendations for your company.

    Contact us

    FAQ