Product shape
Five product areas on one context layer. The official demo walks them in this order.
Screens are 33 frames from the 30 Jan 2026 official demo, with sample data. A generated app is an example output, not a fixed SymphonyAI module.
Foundation: connect, verify, standardise
Raw sources become a published, reusable model before any app runs on them.
Connector catalog
- Card per source: OPC UA/DA, Modbus, ODBC, Aspen InfoPlus.21, InfluxDB, Kafka, Azure IoT Hub, Kinesis, MQTT, OSIsoft, SFTP.
- Namespaces keep each source's own asset hierarchy, down to condenser, turbine and KPI.
P&ID to verified plant context
- Upload a drawing, pick an extraction model, watch each page move through Analyzing.
- Workbench: entity list with category and confidence filters, detections drawn on the drawing, a count such as 61 / 62 verified.
- A person confirms every entity before it reaches the asset model.
Canonical types
Types such as MaintenanceWorkOrder, QCReport and OperatorLog carry mandatory fields and live as Draft until someone presses Publish.
Four sites, one type
A mapping canvas joins the schemas of four plants to one canonical type. Powerful, but the crossing lines give no focus on conflicts or unmapped fields.
Diagnose: from a moved KPI to a cause
The graph supplies relations; impact analysis turns a marked time window into ranked causal paths.
Impact analysis
- Mark up to seven days on a KPI such as Net Power Export, then compare three critical paths.
- Each path shows its contribution, likely causes and recommendations.
- The flow ends at a recommendation. No approval, owner or outcome check appears.
Knowledge graph
Enterprise, region, plant and unit as nodes. The side panel holds documents, trend, alerts, similar and connected assets. Copilot sits in the corner.
Model deploy gate
Anomaly trend with alert markers next to Retrain, Retest and Use in Deployment. Evidence sits beside the decision to promote.
Build: agent flows and generated apps
Flows orchestrates reasoning, tools and people. Forge writes editable application code.
IRIS Flows
A KPI root-cause template runs from inputs through LLM, tool and code nodes to a PDF and an email. Resources (APIs, datasets, Trino, LLMs) are bound per node.
IRIS Forge
"Build an app to monitor asset health" yields a project with Code, Diff, Preview, terminal and a Deploy button. Output is software you can inspect.
Operate: generated role apps
The demo shows apps for a beverage CIP line, HVAC service and power generation.
CIP runtime
- Current step, 10 of 27 steps done, 3 warnings, live metrics.
- Step chain with a Completed, In Progress, Warning, Error, Pending legend. Pick a step for expected vs actual duration and live flow, temperature and conductivity against targets.
Alert queue
Open, In process, History. Each alert shows location, affected systems, impact, recommended actions and event history, with Assign and Start.
Merit-order recommendations
Current vs optimal load and cost impact per unit, with an action such as "Increase load". No approve, owner or result.
Copilot in context
- A drawer over the current screen answers "summary of active alerts" with a table by severity and a written summary.
- Show AI's work, feedback, regenerate, export to PDF.
Analysis
Fact = from SymphonyAI's site, demo or press. View = our assessment.
Platform Fact
- IRIS Foundry is an industrial DataOps platform on Azure that feeds Microsoft Fabric.
- Areas: Foundry, P&ID ingestion, Cortex knowledge graph, ML Studio, IRIS Flows, IRIS Forge, Workspace, Connected Worker, Copilot.
- Clear lifecycle states: Draft and Published types, model runs promoted to deployment, apps deployed through a pipeline.
AI Fact
- Model registry holds anomaly, causal, mapping and Azure OpenAI models.
- Vision extraction on P&IDs, LLM agent nodes in Flows, code generation in Forge, Copilot over current context.
- Novelis began with rule alerts and added ML alerts over 32 plants to build trust (vendor-reported).
Moat View
- Real: breadth of connectors and governed context; one stack from data to generated apps; big-company credibility and capital.
- Thin: home page follows product taxonomy instead of the job; dense dashboards with repeated filters; UI differs across modules and generations.
Gaps View
- Recommendations never show approval, execution or verified outcome.
- Error, permission, stale-data and rollback states are mostly missing from public screens.
- "Unverified" is styled green, which reads as safe.
Market Fact
- Industrial cases named: GSK (connected worker), Vandemoortele (paperless shifts), Novelis (predictive maintenance), Nippon Gases (asset intelligence). Results are vendor-reported.
- Demo domains: power generation, beverage CIP, HVAC service, refining.
For Minder View
- Build now: versioned workflow and field definitions; step runtime; flow and table views of one event log; issue queue with owner and age; decision packet with approval and outcome.
- Later: connector studio, generic agent resource manager, prompt-to-code apps, ML registry UI, enterprise dashboards.
SymphonyAI vs Minder
Where the two products start, and where they meet.
| SymphonyAI | Minder | |
|---|---|---|
| Starting point | Existing OT and IT systems, drawings, CMMS | First-party event capture from workers and signals; legacy systems as adapters |
| Primary user | Data and OT engineers, reliability engineers, executives | Frontline workers and managers on one event log |
| Context model | Namespaces, canonical types, knowledge graph | Plant model and operational graph |
| AI stack | Graph analytics, ML Studio, agent flows, code generation, Copilot | Agent loop over the operational graph via read-only Trino, cited answers |
| Autonomy | Recommend; human approval stated in Flows, not shown in apps | Read-only and advisory in production; humans approve consequential actions |
| Time to first value | Platform build-out before apps | First app live in month 1; Kasa live in six weeks |
| Geography | Global, multi-vertical | Vietnam proving ground, Europe next |
Sources
- Product screens: frames from the official IRIS Foundry demo, 30 Jan 2026, plus IRIS Foundry demo page
- Product pages: IRIS Foundry, P&ID ingestion, Unified Namespace, Knowledge Graph, ML Studio, IRIS Flows, IRIS Forge, Connected Worker
- SIP/CIP on IRIS Foundry
- Cases: GSK, Vandemoortele, Novelis, Nippon Gases
- Company: IRIS Foundry launch, BusinessWire, Mar 2025 (team size), YesPress profile (revenue run rate)
Full research: UI deep dive, 33 frames, 11 flows and SymphonyAI × Ethon screen atlas, 304 fragments. Revenue and headcount come from secondary sources and vary by date. Case outcomes are vendor-reported.