HighByte vs Cognite: Site DataOps Hub or Cloud Data Platform?

This pair usually reaches one shortlist through a group data team: plants want clean, named machine data near the line, head office wants one queryable model of every site and asset. HighByte speaks mainly to the first wish, Cognite to the second.

Fabrico builds cloud OEE software and a light MES, and resells neither vendor. Facts about HighByte are taken from highbyte.com and its user guide, facts about Cognite from cognite.com and docs.cognite.com, as of 13.11.2026.

HighByte vs Cognite at a glance

Each cell condenses a vendor page that is linked further down; all were read on 13.11.2026.

What HighByte Intelligence Hub is for, per HighByte

HighByte's Intelligence Hub product page presents an Industrial DataOps solution built for data modelling, orchestration and governance, and says it was designed for Operational Technology (OT) teams. As of 13.11.2026:

• Location. The page names the edge first, then plant data centres or the cloud, on bare metal, a virtual machine or a container.
• Connections. The user guide's connections section lists types for reading and sending data, among them OPC UA TCP, Modbus TCP, MQTT, Sparkplug, SQL, historians, Kafka, Snowflake and REST Client. How the hub compares with a classic OPC server is in our HighByte vs Kepware comparison.
• Models. In the guide, a model is a collection of attributes, each with a unique name and data type, and models can nest sub models. The product page adds template import and export to repeat a dataset across assets.
• Pipelines. HighByte's pipeline guide has triggers injecting events and stages that perform I/O or transformations on them.

What Cognite Data Fusion is for, per Cognite

Cognite's documentation defines Cognite Data Fusion as an industrial DataOps platform that streams your industrial data into a structured, flexible and contextualised knowledge graph. The Cognite homepage speaks to operations teams, IT teams, digital transformation leaders and executives. As of 13.11.2026:

• Data in. The integration docs describe prebuilt or custom extractors for industrial systems, enterprise applications and cloud sources, and transformations that reshape data to fit target data models. The OPC UA extractor docs say the extractor reads time series, events and asset information over OPC UA.
• Data models. The data modelling docs call a data model an abstract representation of real-world entities, used to build a structured knowledge graph. The core data model supplies standard building blocks such as CogniteAsset, CogniteTimeSeries, CogniteEquipment and CogniteActivity, which Cognite's industry models extend.
• Contextualisation. Per the contextualisation docs, it links time series, files and diagram tags to assets (functional locations) in the knowledge graph, with rules and machine learning for matching.
• Location. The clusters and regions page lets your cloud provider decide the cluster options, across Microsoft Azure, Amazon Web Services and Google Cloud Platform, multi-tenant or dedicated.
• AI. The Atlas AI docs cover industrial AI agents that query the knowledge graph through language models, skills and tools.

HighByte vs Cognite, job by job

Reaching machines. Both name OPC UA, which the OPC Foundation describes as a platform independent service-oriented architecture. HighByte reads it at the plant; Cognite's extractor reads a server and writes to a cloud project.

Shaping data. HighByte's shape is a model per asset type, filled by instances and repeated by templates. Cognite's shape is a data model and knowledge graph that starts from its core model. Decide whether the line model is drawn at the plant and copied up, or drawn centrally and mapped down.

Adding context. HighByte adds it when a model joins raw tags into a named asset; Cognite links time series, files and drawing tags to assets across the company's data.

A shared spine. ISA describes ISA-95 as standards aimed at integrating logistics systems with manufacturing control systems, and says it deals mainly with the interface between levels 3 and 4. Its site, area and line levels make a common spine for both tools (our ISA-95 guide).

Match the first job

If one project must go first, pick the job that blocks people today.

Where OEE and stop causes fit

A HighByte model or a Cognite time series can carry a filler's run state and count. Neither value explains why the filler stopped, how long the shift was meant to run, or how fast that product should run.

For a group, the definitions are harder than the transport. Planned time is one: if one plant books a planned changeover as planned time and another as a stop, head office compares two numbers under one name, as our Kepware vs Litmus comparison works through. The ideal rate is the other, and the example below shows it. Fix the rules once, apply them where OEE is calculated, and send them up with the figure. See our OEE calculation guide and OPC UA for OEE.

Where Fabrico fits for a data-platform buyer

Fabrico does not sell DataOps software, data platforms, SCADA, HMI or edge gateways. Fabrico is OEE software in the cloud plus light MES: beside a hub and a platform, it is where the shift plan, ideal rates and stop causes live.

Signals. Fabrico reads run, stop and count from PLCs, sensors or counters; our connectivity page lists OPC UA, Modbus, MQTT and PROFINET among its protocols. For manual steps or machines without a usable signal, the camera add-on counts output only, with stop-reason suggestion on the Roadmap; its video is streamed on request and not stored. More on the camera add-on page and in computer vision for FMCG manufacturing.

Reasons. Operators pick a reason from a list on a phone or tablet beside the machine. OEE, downtime and losses show by line and shift on web, iOS and Android, as our MES and OEE overview sets out.

Export. Our REST API, webhooks and Excel or CSV files let a data team load OEE, stop causes and planned minutes into its own pipelines. We claim no named connector to either vendor, so prove the route in a pilot.

Curious how one ideal-rate rule looks across two lines? Bring the rates each site uses today to a Fabrico demo.

Worked example: one shift, two ideal rates

Sample figures for an invented dairy plant, not a customer and not a product screen. The site team reports OEE against the rate its line has demonstrated; the corporate data platform rebuilds it from the same counts and stop minutes against the nameplate rate, and gets less.

A yoghurt cup filler runs a 490 minute shift with 30 minutes of planned breaks, so planned time is 460 minutes. Unplanned stops total 40 minutes, so run time is 420 minutes. The filler counts 18,420 pots of the 150 g cup; the checkweigher rejects 180, so 18,240 are good. The machine plate says 60 pots a minute; the best rate the site has held for that cup over a full shift is 48 a minute.

Site figure, demonstrated rate of 48 a minute. Availability 420 / 460 = 91.3%. The ideal for 420 minutes is 420 x 48 = 20,160 pots, so performance is 18,420 / 20,160 = 91.4%. Quality 18,240 / 18,420 = 99.0%. OEE is 91.3% x 91.4% x 99.0% = 82.6%, which matches 18,240 of 22,080 possible.

Corporate figure, nameplate rate of 60 a minute. Availability and quality stay at 91.3% and 99.0%. The ideal for 420 minutes is 420 x 60 = 25,200 pots, so performance is 18,420 / 25,200 = 73.1%. OEE is 91.3% x 73.1% x 99.0% = 66.1%, which matches 18,240 of 27,600 possible.

The filler did the same work in both rows. The nameplate rate shows the speed the machine was built for; the demonstrated rate shows what this line has proven it can hold. The 16.5 point gap lives in the product list, not in the pipes, so both data layers can carry the counts perfectly and it stays. The cure: one agreed source for ideal rates across the group, one owner for the product list, and the rate used sent up with each figure.

Questions for HighByte and Cognite

Give both the same source list, a sample line model and your OEE rule, then ask:

• Which of our controllers, historians and databases can you read without extra servers?
• Where is the line model designed, and how does a change at one plant reach the rest?
• How would you represent a shift, a planned changeover and a stop, with start, end and cause?
• Where is data held when a plant's internet link drops, and for how long?
• If we run both, which product owns asset names and hierarchy?
• Which route between your two products have you tested, and who supports it?

See also our OEE software guide and our page on edge vs cloud shop floor data.

What people ask next

Is HighByte a competitor to Cognite?

They meet at one word, context. HighByte describes Intelligence Hub as Industrial DataOps software that can run at the edge; Cognite's documentation describes Cognite Data Fusion as an industrial DataOps platform that runs in cloud clusters. Ask both how they see the split.

Can HighByte send data to Cognite Data Fusion?

HighByte's guide lists cloud, database, streaming, MQTT and REST connections, and Cognite's integration docs describe prebuilt and custom extractors. Ask both vendors which route between the two they have tested, and who supports it.

Does either one calculate OEE?

Both can hold raw inputs such as run state and counts once a team models them. The figure also needs the shift's planned time, each product's ideal rate and the operator's stop cause, which usually sit in an OEE application or MES.

Why does our corporate OEE differ from the site figure?

Often the ideal rate differs. If the site measures performance against the rate its line has demonstrated and the group platform uses the nameplate rate, one shift gives two figures: 82.6% and 66.1% in the sample on this page. A different planned-time rule is the other usual cause.

How does Fabrico relate to HighByte and Cognite?

Fabrico sells OEE software and light MES, not a DataOps hub or a data platform. It keeps the shift plan, ideal rates and operator stop causes, uses the camera add-on to count output at steps without a usable signal, and hands data on by REST API, webhooks or Excel and CSV files.

Want one OEE rule with operator stop causes across your plants? Book a demo.