# The New Industrial AI Stack > The industrial tech stack is getting new layers: DataOps, knowledge graphs, MCP, agents. See where Fuuz fits and what buyers should know before building it. Source: https://www.fuuz.com/platform/industrial-ai-stack The industrial AI stack After IMTS 2026, IoT Analytics described a new, AI-led industrial stack forming above the ISA-95 pyramid: industrial DataOps, a knowledge graph, interfaces such as MCP, agents, and apps where people stay in the loop. Most manufacturers would have to buy and integrate each layer. On Fuuz, most of them are one governed platform. [See where Fuuz fits](#stack) [Talk to an architect](https://www.fuuz.com/contact) Where Fuuz fits ## One Platform Across The Stack Each layer of the new stack, top to bottom, and what it means in Fuuz. LayerFuuzWhat it means in Fuuz 1. **Applications and human-in-the-loop** Where people review, approve and act *Native* Operator, supervisor and engineering apps built on Fuuz, with approvals and electronic records in the workflow. 2. **Agents** Orchestrators and specialist agents *Your agents* Your corporate agents, such as Claude, GPT, Gemini or your own, reach every part of Fuuz through MCP. Agent calls can also run as steps in data flows, traced and audited. 3. **Interfaces** How agents reach software and data *Native* MCP tools, pre-built or your own, one GraphQL API for every record and application, and the industrial and enterprise protocols your systems already speak. 4. **Knowledge graph and ontology** A model of assets, processes and relationships *Native* One data model with an ORM that turns your data into a knowledge graph across the plant, ERP, PLM and CRM. 5. **Systems of record** ERP, MES, PLM, CMMS and more *Run or connect* Run MES, WMS, QMS and CMMS on Fuuz, or keep the ones you have, with history stored in Fuuz for most use cases. ERP and PLM connect two ways. 6. **Industrial DataOps** Connectivity and contextualization *Native* Connectors, data flows and an enterprise Unified Namespace that persists state and context, not just messages. 7. **Software-defined edge** Compute and connectivity near the machine *Native: Fuuz Gateway* The Fuuz Gateway is a native part of Fuuz, not a third-party add-on: a software-defined gateway at each site with edge connections, data flows, a local UNS and local agents, outbound-only. 8. **Physical process** Machines, robots and production lines *Connected* PLCs, sensors, robots, scales and scanners connect through the Fuuz Gateway. Nothing on the floor is replaced. Native in Fuuz, including the GatewayRun on Fuuz or connectConnected Layers adapted from [IoT Analytics, “IMTS 2026 recap: A new industrial AI tech stack is forming” (October 2026)](https://iot-analytics.com/industrial-ai-tech-stack/). Fuuz mapping by Fuuz. Why it matters ## Context Is The New Advantage The clearest signal from IMTS 2026 was where value is moving: away from tools that only move data, toward platforms that own its context. Vendors told IoT Analytics that the data model and the knowledge graph are what set them apart, and even hyperscalers now argue for connecting and contextualizing data where it lives instead of pooling it in one lake. That is the layer Fuuz was built around. Data is connected, persisted and modeled once, and every app, agent and API works from that same model, under the same security. And in Fuuz, that model is more than a map of your data: it is the backend your applications run on. - One model of assets, orders, lots and processes - Agents follow typed relationships instead of guessing - Apps, agents and integrations share one security model The stack, two ways **Assembled** A DataOps tool, a graph database, an MCP gateway, an agent platform and apps, each with its own data and security. **On Fuuz** Those layers on one platform, one model and one security model, including the native Fuuz Gateway at the edge, with your systems of record run on Fuuz or connected. **Result** Fewer seams, less integration upkeep, and AI that starts from data you can trust. Not just the graph ## Fuuz Is Your Application Backend Most knowledge graphs sit beside the systems that run the business, as one more integration point. Fuuz is where your applications actually run. That is why its context is the most valuable in the stack. A graph beside your systems ERPMESWMSQMS **Sync jobs and mappings** **Knowledge graph** A copy, as current as the last sync - Every system upgrade can break a mapping - Context drifts from what really happened - Agents read a copy and write back somewhere else Fuuz as your application backend MESWMSQMSCMMS **Apps read and write the model directly** **One model and knowledge graph** Where the work happens, so it is current ERPPLMMachines*Connected two ways* - Context is created as the work is done - Change the model once, every app follows - Agents act through the same rules and security Why it matters ## Context That Stays True ### Current because the work happens there When production, inventory, quality and maintenance run on Fuuz, every transaction lands in the model as it happens. The context is not copied from somewhere else, so it does not go stale between syncs. ### Survives the next upgrade A graph fed by integrations has to be rebuilt every time a source system changes. In Fuuz, apps are built on the model, so the context stays intact as systems around it are upgraded, replaced or added. ### Context agents can act on Your agents do not just read a picture of the operation. They work through the same data model, business rules, permissions and audit trail as your apps, so an action taken by an agent is as governed as one taken by a person. ### One model, many apps Extend the model once and every app, dashboard, API and agent can use it. Teams stop rebuilding the same context for each new tool. Deterministic by design ## Intelligent Agents. Deterministic Outcomes Agents are powerful, but on their own they guess. In Fuuz, agents do the reasoning and Fuuz flows do the work: typed, governed and repeatable, at the edge or in the cloud. That is what makes agents safe for any industrial scenario. Your agent **Decides what to do** Claude, GPT, Gemini or your ownTyped request Role-based access **Checks who is asking** Only the data and tools that user may useAllowed Fuuz flow, edge or cloud **Decides how, the same way every time** Typed inputs, typed outputs, your business rulesResult Headless or headed **Runs on its own, or waits for a person** Approvals where you want them, audit everywhereHow it works ## Agents Act Only The Way You Allow ### Same input, same answer When a tool is typed, a Fuuz flow returns the same result for the same input, every time. The agent chooses what to ask. The flow decides how it is done, by your rules. ### Only the context you give Agents can act only on the data, tools and context you expose to them. Nothing else in the platform is visible to them. ### Role-based access on every call Every tool call runs under the signed-in user's role and permissions, with tool policies on top, and every change is recorded in the audit log. ### At the edge or in the cloud Flows run where the work is: on the Fuuz Gateway next to the machine, or in the cloud. Agents get the same deterministic behavior in either place. ### Headless or headed Run fully automated, event-triggered operations, or keep a person in the loop to review and approve inside a Fuuz app. You choose per process. ### Safe for real operations Because the outcome is governed by typed flows and permissions, agents can take on real industrial work, from a quality hold to a work order, without guessing. What buyers need to know ## Before You Build The Stack, Ask These Questions ### Watch for vendor-locked agents Some vendors build in their own agent with hard-coded prompts. Every one adds another silo. Ask whether your corporate agents can reach the whole platform, using your company's knowledge instead of the vendor's. ### Count the integrations, not the logos Eight layers bought from eight vendors means eight data copies, eight security models and dozens of integrations to keep alive. Every seam is where context gets lost and projects stall. ### Look behind the MCP server An MCP server is only as good as what sits under it. Ask whether it exposes a real data model and relationships, whether agents can write back or only read, and whether every call runs under the user's own permissions. ### Decide who owns the knowledge graph Context is becoming the most valuable layer in the stack. Know whether your model of assets, orders, lots and processes lives with you, or inside a vendor's product or a hyperscaler's service. ### Keep people in the loop, on purpose Engineers will not trust an agent they cannot check. Plan approvals, traceability and audit from the start, so every agent action can be reviewed, explained and reversed. ### Bring IT and security in early Shop-floor software decisions increasingly sit with the CIO. Identity, access, data residency and change control decide whether a pilot ever reaches a second plant. ### Build on standards that exist today MCP, APIs and proven industrial protocols work today. AI-specific standards such as A2A (agent to agent) and MHS (hardware access for models) are still early. Build on what works now, on a platform that can take on new standards without a rebuild. ### Ask how AI access is licensed Vendors differ: some include AI and MCP access, others plan to charge for it. Get it in writing before your agents depend on it. ### Fix the data foundation first Most industrial AI projects stall on scattered, uncontextualized data. Connect, persist and model it once, and every layer above it gets easier. A practical path ## How Manufacturers Get There On Fuuz Step 1 ### Connect and model Connect machines and enterprise systems through the native Fuuz Gateway and connectors, and define your data model once. That model becomes your knowledge graph. Step 2 ### Run or connect systems of record Run MES, WMS, quality and maintenance on Fuuz, or keep what works. Either way, every app works from the same model. Step 3 ### Expose governed tools Turn data, flows and connected systems into MCP tools and GraphQL queries, under role-based access and tool policies. Step 4 ### Put agents to work, with people approving Embed agents in flows and apps, keep humans approving the decisions that matter, and audit every change. Questions ## The Industrial AI Stack, Answered What is the new industrial AI stack?It is a way to describe the layers manufacturers now need above and beside the classic ISA-95 pyramid: industrial DataOps, a knowledge graph and ontology, interfaces such as MCP, AI agents, and applications where people stay in the loop. The framework on this page is adapted from [IoT Analytics' IMTS 2026 recap](https://iot-analytics.com/industrial-ai-tech-stack/). Does Fuuz replace ISA-95?No. ISA-95 still describes how plants and enterprise systems are organized, and Fuuz works within it. The new layers sit above and beside that foundation, and Fuuz provides most of them on one platform. Does Fuuz replace my ERP?No. Fuuz connects two ways to ERP, PLM and other systems of record. You can run MES, WMS, quality and maintenance on Fuuz, or keep the systems you have and connect them. Does Fuuz support MCP?Yes. Fuuz includes pre-built MCP tools, and you can build your own on top of any connection. Agents connect through OAuth, every tool call runs under the signed-in user's permissions, and every change is recorded in the audit log. How is Fuuz different from a standalone knowledge graph?A standalone graph is fed by integrations and is only as current as its last sync, and it has to be remapped whenever a source system changes. Fuuz is also the application backend: MES, WMS, quality, maintenance and the apps you build read and write the model directly. The context is created as the work is done, and it stays intact as other systems are upgraded. Is Fuuz a knowledge graph?Fuuz has a data model with an ORM layer from day one. That ORM turns your data into a knowledge graph that spans the plant and connected systems such as ERP, PLM and CRM, so agents follow typed relationships instead of guessing. Will an agent give the same answer every time?When the work is done through a typed Fuuz flow, yes. The agent decides what to ask, and the flow returns the same result for the same input every time, by your business rules, at the edge or in the cloud. Agents can act only on the data and tools you expose, under role-based access, headless or with a person approving. Where do agents run?Both at the edge and in the cloud. Local agents can connect to the Fuuz Gateway at a site, and cloud agents connect with OAuth so users get secure access under their own permissions. Get started ## Build The New Stack On One Platform Talk to our team about where your stack is today, which layers you already have, and what Fuuz can take on. [Talk to an architect](https://www.fuuz.com/contact) [See AI in Fuuz](https://www.fuuz.com/platform/ai) Prefer to talk? Call [1-877-801-FUUZ (3889)](tel:18778013889)