Build A Better UNS With AI-Powered DataOps
Timebase Atlas is the map your people, your AI, and your applications need to understand how your plant actually operates
.png)
.png)
.png)
What Is Atlas?
Existing Systems Store.
Atlas Means.
Atlas is a new layer that sits above these stores and layers meaning on the data they already hold — what the values are, how the equipment relates, and why it matters.
The operational knowledge that today lives in the senior engineer's head and nowhere else.
.png)
The Anatomy of a Knowledge Model
A Backend Built For The Problem
Owning the knowledge graph means no per-node licensing drag, no external cluster to stand up before the first model exists, and an install footprint light enough to run at the edge of a single line or across an enterprise. The same engine that keeps a small model fast keeps an enterprise model with millions of relationships responsive enough for an agent to reason on in real time.
relationships are first-class citizens
- that the feed only occurs during fermentation
- the flow path it runs through
- the conditions under which it is valid
- the date it was commissioned
The flexibility is deliberate. You define the relationship types your operation actually uses, with the properties that matter to you, rather than bending your world to fit a fixed set of hierarchy rules. Two plants can model the same equipment differently because they run differently, and both models are valid.
libraries & governance
Some libraries ship with Atlas and encode published standards, so nobody designs them from scratch: ISA-95 for the enterprise, plant, area, and work-center hierarchy, ISA-88 for batch, recipe, procedure, and phase, ISA-5.1 for parsing instrument tags such as FIC-101 or PSH-202 into structured attributes, ISA-18.2 for alarm management, ISO 15926 for oil and gas asset lifecycle, and CESMII Smart Manufacturing Profiles. You import the standard, and it becomes live ontology in your model.
You also create your own, and this is the bigger idea. A control-module library of valves, motors, and PID loops, a corporate equipment standard, a recipe framework specific to your products. When an organization authors its own libraries, it takes governance over its model: the vocabulary is defined once, owned centrally, and every domain that consumes the library models under it.
Standards give you a proven starting point; your own libraries are how your organization governs what things mean. Once built, a library is reusable across every model that needs it, which is how a standard actually propagates across sites instead of being re-typed at each one.
Domains and how they federate
It is where a team actually does its modeling, and it consumes libraries to do so: a process-engineering domain might pull in ISA-88 and a recipe library, a maintenance domain pulls in an equipment-standard library and ISA-18.2, and each team models its own slice of the operation using the vocabulary those libraries provide. They do not need to understand the full enterprise model to contribute correctly to their piece of it.
When domains are linked, the linkages are explicit relationships in the graph, not assumptions built into a shared hierarchy that breaks when someone at Site 7 names something differently than Site 3. The power of Atlas as a modeling tool is the organizations to standardize while remain flexible.
How the model Is built, used, & governed
.png)
Gage is the agent. It ships with a deep set of prebuilt skills and it can orchestrate subagents, so a modeling session is not one model answering one prompt, it is Gage coordinating specialized work against your model. Gage and its shipped skills can be used to author new skills, which are then stored in Atlas and become available to Gage and its subagents from that point on. The operation's modeling discipline compounds inside the platform rather than living in a consultant's head.
Gage is also not confined to Atlas. It becomes your agent orchestration layer across all Timebase components, which means the same agent that builds and reasons on the Knowledge Model reaches the historian and the rest of the stack. One agent, one governed model, working across the entire platform rather than a chat box per product.
.png)
The agent partners with the domain expert. The agent knows how to model; the engineer knows what to model. It asks questions, proposes structure, and externalizes the tacit knowledge that has never had a home in any system. It does not generate a model on its own and ask for a rubber stamp. Domain experts stay the source of truth, and the agent is leverage on their expertise rather than a replacement for it. This is the difference between a tool that helps a knowledgeable engineer move faster and one that produces plausible structure nobody trusts.
.png)
Skills are packaged modeling know-how: how to decompose a process line, how to represent a recipe, how to handle CIP cycles, how to elicit knowledge from a senior operator. They are decades of manufacturing modeling discipline made executable, and they are not a fixed set: your team builds new ones through Gage and keeps them in Atlas. A competitor who clones the graph database does not clone the skills, which is a large part of why the model actually gets built in Atlas rather than started and abandoned.
.png)
The System Domain is Atlas's memory of the business: the operation's own philosophy, naming conventions, alarm philosophy, engineering standards, and change-management procedures, captured as a queryable model. It is the plant's operational constitution, and because it is a model rather than a wiki, Gage and every subagent work within it rather than around it. A strong demo moment: ask the agent a modeling question and watch it answer in the operation's own standards.
.png)
Models are Git-backed. Every change to the ontology, the instances, and the relationships is versioned, which means a model has history, branches, review, and rollback the same way source code does. For a governance layer this is not a convenience, it is the mechanism: you can answer how the model looked during Batch 47, who changed a definition and when, and you can propose and review a structural change before it goes live. Governance stops being a policy document and becomes an enforced property of the platform.
.png)
Once built, the model is available to any agent that can call REST, through the MCP server and the .NET Agent SDK, the same pattern the Timebase Historian MCP server already uses. Progressive context traversal means an agent pulls only the relevant subgraph rather than the whole model, which keeps context windows clean and responses fast and reliable. The model is large; the context window is not, and the harness is built around that fact.
.png)
Everything that crosses the boundary
.png)
Atlas connects to every shape and source of data a real plant produces. The point of connecting a source is not to copy it: Atlas relates across sources and knows where the authoritative record lives without becoming the owner of it.
.png)
Bridge is the outbound mirror of Connector. It publishes data out of Atlas to downstream sinks: BI tools, external systems, other brokers, and custom consumers.
Atlas can publish Sparkplug, and Timebase Historian is a full-spec Sparkplug subscriber, so the model can feed the same UNS transport it draws from. This is what lets Atlas sit as the semantic layer above the namespace while still speaking the namespace's native language in both directions.
