Questions before Koios goes in? Start here.

At least 2 cores, 4 GB of memory and 20 GB of disk. For a first site we suggest 4 cores, 8 GB and 100 GB on an SSD, which comfortably runs a few hundred tags with a model or two. Larger sites go up to 8 cores, 32 GB and 500 GB.

System requirements →Needs a free account

Ubuntu Server 22.04, 24.04 or 26.04 LTS, on 64-bit Intel and AMD (amd64) or ARM (arm64). Koios runs as a single Docker container, and the installer brings Docker with it, so there's nothing else to set up.

System requirements →Needs a free account

Yes to all three. Hyper-V, VMware, Proxmox and VirtualBox are supported, and a VM is the most common choice. A fanless industrial PC in the cabinet works well for loops that can't wait on a network. A Raspberry Pi 5 running from an SSD suits a modest tag count.

Running Koios on a VM →Needs a free account

No. Many sites run fully air-gapped. The installer carries everything, you can bring models in as files, and the license activates offline by carrying a small file to any computer that has a connection.

Installing offline →Needs a free account

Koios runs on an Ubuntu Server machine of its own, a virtual machine or a physical computer, and your team uses it from a browser, so nothing is installed on the operators' machines. Once Ubuntu is on, one installer does the rest, with or without internet. Every command is written out, and no Linux experience is needed.

Installing Koios →Needs a free account

Yes. Each one keeps its own data, users and license, which suits separate lines, sites or test setups on shared hardware. Size the server for all of them together.

Running more than one →Needs a free account

OPC UA, with server discovery and signed or encrypted connections. EtherNet/IP for Logix, SLC 500 and MicroLogix, and generic CIP devices from their EDS file. Modbus TCP, Microsoft SQL, REST and SOAP. It reads tags from all of them, and writes back where you allow it.

Protocols →Needs a free account

Today, protocols are built into Koios, and for something they don't cover, a custom Python component can act as the adapter on the canvas. Koios 2.0 brings protocol packages: drivers you add, version and update on their own. If there's a protocol you need, tell us. It shapes what we build next.

Only where you let it. A device tag is read-only until you make it writable. Model outputs can be clamped to a range, a prediction that isn't a real number writes nothing, and Koios flags it if two sources try to write the same tag.

Writing is its own permission, separate from configuring, and every write lands in a tamper-evident audit log. Your native controller stays underneath, ready to take the loop back.

Tags and writes →Needs a free account

For OPC UA, yes. A device set groups a primary OPC UA server with its standbys, such as a redundant pair, and Koios switches between them automatically if the primary drops.

Device sets →Needs a free account

Dashboards of live values and the state of each device and model, set up for a process, an area or a role. Trends chart any tags over time, live or looking back, and export what they show. It all runs in the browser.

Dashboards →Needs a free account

Yes. The events feed records what Koios is doing: a device dropping off, a tag going into alarm, a model that fails to run. Your team acknowledges and mutes them as they're handled, and the audit log keeps a separate, permanent record of who changed what.

Events →Needs a free account

Yes. A calculated tag works from a formula over other tags, such as a unit conversion, an average of several sensors, a smoothed reading or an alarm threshold, and updates as soon as they do. Value mapping turns a device's raw codes into states your team can read.

Calculated tags →Needs a free account

Yes. Export your tags to a spreadsheet, add or edit rows, and import them back, on the same server or another one.

Importing and exporting tags →Needs a free account

Yes, through its API. Each system gets credentials of its own, with only the permissions you give it, and you can switch it off at any time. From Python, the Koios client on PyPI does the work for you.

API clients →Needs a free account

ONNX and TensorFlow Lite, on the CPU. Train in PyTorch, TensorFlow, scikit-learn or Ronin, export the file, and bind its inputs and outputs to tags. A new version swaps in without touching the bindings.

Model files →Needs a free account

No. Koios runs models from any tool that exports ONNX or TensorFlow Lite. Ronin is there if you'd rather train and grade models on your own history without a data science team.

Yes. Write components in your own IDE with the Component Builder from PyPI: MPC, optimization, a soft sensor, a state machine. Package them as a library and wire them onto the canvas next to your tags and models. Bring the packages your code depends on too, and each environment keeps its own versions.

Custom components →Needs a free account

Operator, Engineer and Administrator roles come built in, and you can make your own. HTTPS is on from the first start, and you can swap in your own certificate. A tamper-evident audit log records sign-ins, configuration changes and writes, and you can export it.

Roles and permissions →Needs a free account

Not from Koios. It runs on your network, keeps its history on its own disk, and needs no connection back to us. Ronin and Curtis are cloud products, and plant data reaches them only as the historian exports you choose to upload.

As long as you choose, from a week up to ten years or with no limit at all. History is compressed as it's stored, and you can tune that per tag.

Retention →Needs a free account

Koios backs itself up while it runs, daily or weekly, either the configuration alone or with history. Restoring works from a backup on the server or an uploaded file, and an older backup is brought up to the current version as it restores.

Backup and restore →Needs a free account

Per site, activated on the machine Koios runs on, online or offline. Each license runs for a term you choose, and you manage licenses and activations in the customer portal. Pricing depends on the deployment, so talk to us about yours.

Licensing →Needs a free account

No. Koios installs and starts without one, so you can set it up first. Devices and models start running once it's licensed.

Koios keeps running for a grace period of up to 30 days, with a notice on every page so it doesn't catch anyone out. Renew in the portal, upload the new license file, and nothing stops.

Licensing problems →Needs a free account

Yes. A license is tied to the machine it was activated on, so on a new server or VM you activate it again, the same way as the first time, and a backup carries your configuration and history across. If the license has no activation to spare, support can free the old machine's.

Licensing →Needs a free account

Yes. Ask us for a 30-day trial and try it with your own equipment and data. Questions are welcome before and during.

A new version is a new container image. Connected sites pull it; offline sites download the image archive from the portal and carry it in. Your configuration and history stay put, and Koios saves a copy of its database before the first start on a new version.

Updating Koios →Needs a free account

Ronin for training models on your own process history, and Curtis, the assistant that knows your projects and your account. Your administrator assigns seats in the portal and can move them as the team changes.

As a file. Ronin exports an ONNX model, you upload it to Koios and bind it to your tags, and it keeps running with no line back to the cloud.

In the customer portal, open to anyone with an Ai-OPs account. Making one is free, and customers already have one. Once you're signed in, every docs link on this page opens its page.

Sign in to read the docs →

Yes. They cover each released version of Koios, with a picker to switch between them. The same docs ship inside Koios itself, so a site with no internet has them too.