Koios 1.0

Koios 1.0 runs your own Python logic and calculated tags beside your devices and models, and rebuilds the interface around dashboards, trends and tables you can act on in bulk. Backup and restore, network diagnostics, users and roles all move into the browser, and every installation generates its own secrets.

• Before you upgrade: add the secrets volume before the first start
• Your own Python logic runs inside Koios: components wired to tags and models, at the scan rate you choose
• Expression tags calculate values from live data: formulas over tags, devices and models, with no code
• Models run in scan groups, on demand or from memory: synchronized runs, fresh reads, and fast control loops
• Dashboards and trends are yours to arrange: widgets on a grid, and trends styled your way
• Backup and restore run from the browser: scheduled or on demand, with or without history
• What else changed: devices and tags, models, lists, accounts and security, system and installation

Before you upgrade

1. Add the koios_secrets volume to your service file or docker run command before starting 1.0.0. Without it the container will not start. See Running Koios as a Service.
2. Existing credentials carry over: the first start writes the values your database already uses to the new volume.
3. Lookup tables become value-mapping rules. Each key and value becomes an exact-match rule, and a * entry becomes a catch-all rule. See Value Mapping.
4. The first start backs up the database and then migrates it, which can take a few minutes on a large system.
5. An older version will not start against the migrated database. To roll back, restore a backup taken before the upgrade. See Updating Koios.

Step-by-step instructions: Migrating to v1.0.0.

Your own Python logic runs inside Koios

Components are Python packages that run inside Koios beside your devices and models. Wire their inputs and outputs to tags, models or other components on a canvas, and group them into environments, each running at its own scan rate from 0.1 seconds to one hour. Use them for control algorithms, data transformations, state machines and custom protocol adapters. Upload libraries of components as .kcl files and manage their versions, or write your own with the Component Builder library, which gives you typed inputs, outputs and configuration fields. Every execution and configuration change is recorded.

See Components.

Expression tags calculate values from live data

An expression tag computes its value from a formula instead of reading a device. A formula can combine values from tags, devices and AI models, with arithmetic, comparisons, conditions and boolean logic. It is evaluated continuously, and its result is stored, trended and shown to the decimal places you set, like any other tag, so it can feed a dashboard, a model input or another expression. Formulas run in a restricted evaluator, so an expression cannot execute arbitrary code.

See Expression Tags.

Models run in scan groups, on demand or from memory

A scan group runs related models together on one schedule, sharing on-demand settings, so every model in it works from the same snapshot of fresh data. A model or scan group can also run on demand: it asks its devices for fresh readings before each inference and writes its outputs back straight after, closing a control loop within one scan cycle. For the fastest loops, memory-only mode skips the time-series database on every cycle while the model still publishes live values for dashboards and other consumers.

See Scan Groups and On-Demand Inference.

Dashboards and trends are yours to arrange

The interface is rebuilt as a single-page application: pages switch without reloading, live values update faster, layouts fit desktops, tablets and wall displays, and every page works from the keyboard. A dashboard holds live widgets on a drag-and-drop grid, in eight types across tags, devices and models. Keep several dashboards for different areas or roles, and mark one to open when you sign in. A trend keeps the styling you give it: a color and axis per trace, several Y axes with their own bounds, line and fill styles, and aggregation by mean, median, minimum, maximum, sum, first or last, with gap filling and CSV export.

See Dashboard and Traces & Axes.

Backup and restore run from the browser

Back up and restore Koios from the System menu, with no command line. A configuration backup captures devices, tags, users, models and settings in seconds; a full backup adds historical data. Schedule backups daily or weekly at 2:00 AM, keeping between 1 and 50, or run one on demand. To restore, upload an archive, review what it contains, and choose whether a full backup brings its history with it. Archives download for keeping off the server.

See Backup & Restore.

What else changed

Devices and tags

• Value mapping is a rules engine: ordered rules match exact values, numeric comparisons, ranges, wildcards or anything, with case-insensitive matching and reverse mapping on write. See Value Mapping.
• A device can be read or written on demand, outside its normal scan cycle. See On-Demand Scanning.
• On-demand requests to an unreachable device back off and recover on their own, so one device cannot hold up the rest. See On-Demand Scanning.
• Tag value changes are pushed to the interface as they happen, so live values update sooner.
• OPC-UA client certificates are managed from the interface. See OPC-UA Certificates.

Models

• Each input binding normalizes as none, min-max (0 to 1), symmetric (−1 to 1) or z-score, from the tag's range or values you enter. See Assigning Bindings.
• A binding can flag an input that changes too fast, by an absolute or percentage threshold over a window, rising, falling or both. See Assigning Bindings.
• A new model file can re-apply the settings in its metadata, such as sample rate, model type and normalization, while keeping your tag assignments. See Managing Model Files.

Lists and trends

• Every table supports bulk enable, disable and delete, CSV import and export, sorting and filtering, with progress shown while a bulk action runs. See The Device List.

Accounts and security

• Each installation generates its own database passwords, session signing keys and service tokens at first start, so no two installations share secrets.
• Users, custom roles and granular permissions are managed in the interface. See Roles & Permissions.
• API clients get their own credentials for connecting other systems. See API Clients.

System

• System Health shows CPU, memory and disk use live, with warning and critical thresholds you set. See System Health.
• Every platform service's status is shown in one place, grouped by category. See Services.
• Service logs stream live in the browser, colored by level, with pause and resume. See Logs.
• The Network page lists the server's interfaces and runs ping, traceroute, DNS lookups and TCP port checks, streaming the results. See Network.
• Retention policies clean up historical data automatically and track storage use. See Data Retention.
• A burst of events, such as enabling hundreds of tags at once, shows as one summary notification instead of flooding the screen. See Events.
• API rate limits are higher and tuned per endpoint, so dashboards with many live values keep updating.
• The full help system ships inside Koios: more than 80 pages, searchable, and available without an internet connection.

Installation

• Images run on x86-64 and ARM64 under the same tag. See System Requirements.
• ENABLE_TLS, HTTP_PORT and HTTPS_PORT set how Koios serves TLS, including turning it off behind a load balancer that terminates it. See Environment Variables.
• Every start checks the database against the version starting, and an older version refuses to run on a newer database. Each upgrade backs up the database before migrating it.

Patch releases

v1.0.1

• Enabling and disabling a device or model in quick succession no longer leaves duplicate device polls or model runs behind.

v1.0.2

• A model's own prediction history records inputs that come from in-memory tags, such as component outputs and expression results.
• A binding change made while the expression evaluator is restarting still takes effect within one refresh cycle.
• Models start after a restore even when their history storage already exists.
• An ONNX file above the IR version the runtime can load is refused at upload with the reason, instead of failing when the model runs. See Managing Model Files.
• Model file metadata is read with koios-model-utils, so malformed metadata is refused at upload instead of leaving blank or wrong settings.
• Long component names no longer overflow the canvas library tray.
• Bindings fed by in-memory tags are labeled as in-memory, not as expressions.
• Exporting every tag with no filter applied works.
• The koios-component-builder package is licensed under Apache 2.0.