Product shape
"Connect, contextualise, simulate, decide, execute, verify", in their own words.
Industry modules: facility ops, predictive maintenance, semiconductor (SemiOps), data centres, traffic and border operations, district heating. Deployment: SaaS, private cloud, on-prem or hybrid.
FactVerse AI Agent
Launched 10 March 2026: "Turn operational evidence into the next approved action."
Analysis in the web app
- Left nav per domain: heat, traffic, vision, SemiOps (cleanroom), production; AI Chat and Alerts on top.
- Here a cleanroom correlation study: temperature vs humidity, R² 0.755, Pearson 0.869, best lag −2 h, and a correlation matrix across airflow, pressure, particles and humidity.
- Tabs for monitoring, prediction, soft sensors and cleanroom simulation.
| MCP pack | Tools | Examples |
|---|---|---|
| Base | 54 | detect_anomaly (z-score, isolation forest, autoencoder), predict_rul, explain_prediction (SHAP), forecast_timeseries, estimate_causal_effect, run_des, optimize_milp, optimize_layout, query_knowledge, extract_maintenance_record, search_documents, calculate_emissions, create_work_order (needs approval) |
| Semiconductor | 16 | SMT OEE, defect Pareto, ISO 14644 cleanroom, chiller efficiency |
| Aviation MRO | 11 | Weibull reliability, ATA fleet statistics, repeat faults |
| Data centre | 10 | Alert diagnosis, dispatch with approval |
| Traffic | 7 | Queue forecasts, lane reconfiguration, staffing |
| Predictive maintenance | 5 | Health, remaining life, maintenance priority |
Scopes per pack (read, compute, write), keys per tenant shown once, scope checks fail closed. Most analytics are classical ML and operations research packaged as tools. The launch said "21 tools"; the docs now list 103.
Minder: the closest public twin of our agent-over-MCP design. Their tool list is a useful checklist of what buyers will expect an industrial agent to do.Twin and field apps
The 3D twin is the hub; Inspector carries alarms into work orders on desktop and mobile.
Predictive maintenance
- Fleet dashboard for a demo fab: 10 tools, alerts, 96% average health.
- A maintenance priority queue per tool (scanner, implanter, filter units, cooling water pump, compressor) with an evidence chain and an "Optimize Schedule" action.
- The nav shows the rest of the suite: cockpit, operations centre, CMMS, vibration, anomalies, advisory inbox, sustainability.
Twin workbench
A cleanroom building with layers for equipment, pipes, safety, HVAC; equipment online rate, alerts and work orders on top; drafts and published revisions of each twin.
Inspector
Command Center in 3D: an air handling unit flagged, linked to its sensor, with device data. Mobile shows hidden pipes in AR, runs offline QR/NFC rounds, and an AI copilot pre-fills work orders.
Training and simulation
The original business, and still where most of the proof is.
Director
No-code authoring of 3D work instructions and training: exploded models, step storyboards, quizzes, one-click translation. Plays on PC, iPad, Apple Vision Pro and Quest; up to 100 people in one session.
Simulator
Heavy-equipment operator training on motion platforms: tower crane, excavator, rotary drill rig. SANY and Sunward use it; there is even a Steam version.
Other lines: Designer (no-code scene and process simulation, NVIDIA Omniverse export), Checklist (inspection app, not updated since July 2024), Robotics (synthetic training data for robots, early access).
Use cases
All values are DataMesh's own; most pages give none.
| Area | What DataMesh does | Claim |
|---|---|---|
| Discrete manufacturing | 1:1 factory twin, AGV and layout simulation, XR training (NIO, EVE Energy, Foxconn) | Training −40-70% (Faurecia); +30% training efficiency (Foxconn) |
| Line planning, machine OEMs | Scenario simulation to sell and plan new lines (Jebsee, Chen Hsong, Gyro) | Twin in 2 weeks; ~18% throughput gain, simulated |
| Predictive maintenance | Data fusion, AI agent, Inspector work orders; Yokogawa sensors | Risk flagged 10-14 days earlier (proof of concept) |
| Semiconductor fabs | Cleanroom ISO 14644, particles, pressure, utilities, chiller efficiency; airflow studies | None |
| AI data centres | Power and cooling fusion, alarm to dispatch, thermal what-ifs, Omniverse DSX | None; one unnamed telecom |
| Buildings and parks | Inspector with Brick ontology, Green Mark evidence | None |
| Energy | Energy-use review, control drift, EnergyPlus scenarios, Scope 1-3 emissions | None |
| District heating | Hydraulic and thermal models, imbalance diagnosis, preheat planning | None |
| Borders and ports | 8-hour queue forecasts, lane reconfiguration, staffing | None |
| Aviation MRO | Reliability analysis tools | Unnamed customer |
| Construction, expressways | BIM to mixed-reality sequences (JTC, Obayashi); MR training (NEXCO-East, 1,000+ trainees a year) | None |
| Bottling maintenance | Director instructions at Swire Coca-Cola | Downtime −1/3, "expected" |
Proof and pricing
Credible logos, few measured results.
Proof Fact
- Faurecia Yancheng: training time −40-70%, equipment time used for training down two thirds.
- Foxconn: training efficiency +30%, spot-check failures −4% (phase 1).
- Yokogawa joint PdM: risk windows 10-14 days earlier (proof of concept).
- Faurecia and EVE Energy plants hold WEF Lighthouse status; the recognition is theirs, DataMesh supported.
- Gartner Tech Innovator for Intelligent Simulation (2025).
Pricing Fact
- Director: $199/month or $1,999/year (1 creator + 3 viewers), buy online.
- Designer: same price; Suite from $229/month.
- Inspector, Simulator, AI Agent, Data Fusion: by quote (data fusion per node or server).
- 30-day trial for the authoring tools.
Analysis
Fact = from DataMesh's site, docs or press. View = our assessment.
Go-to-market Fact
- Channels in Japan: KDDI Business Port, Fujitsu, TIS, Otsuka, Yokogawa; Synnex in Taiwan.
- Partners: NVIDIA (Omniverse council, DSX blueprint), Apple, Microsoft; joined the OpenUSD alliance in 2026.
- Google for Startups Accelerator SEA 2026; KDDI backs expansion into Southeast Asia.
Moat View
- Real: years of 3D and XR tooling; simulation engines; Omniverse alignment; big-factory logos; cheap self-serve entry.
- Thin: the agent is seven months old and mostly packages classical analytics; the "world model" is project services; ops results are unmeasured.
Weak spots View
- Small team across five offices; R&D in Beijing, which some SEA buyers may weigh.
- Product sprawl: 12+ products, HoloLens support frozen, Checklist stale.
- Numbers on their own site disagree: 500+, 400+ or 200+ customers.
For Minder View
- Take: approval-gated write tools, an evidence chain per asset, AI-assisted tag-to-asset mapping (their Scout).
- Win on: Vietnamese-first frontline capture, measured plant results (energy per product), a model trained on the plant's own record.
DataMesh vs Minder
Where the two products start, and where they meet.
| DataMesh | Minder | |
|---|---|---|
| Starting point | A 3D twin from CAD/BIM, bound to live system data | First-party event capture from workers and signals; legacy systems as adapters |
| Primary user | Engineers, facility and maintenance teams, trainers | Frontline workers and managers on one event log |
| Agent | FactVerse AI Agent, 103 MCP tools, approval before writes | Agent over the operational graph via MCP endpoints, cited answers |
| Frontline | Checklists, Inspector rounds and evidence, XR instructions | Workers log tasks, decisions and outcomes in their own language |
| World model | Data + structure + physics solvers; no trained model | Trained on the plant's decision record to predict the effect of a decision |
| Simulation | Strong: 9 what-if engines, airflow, thermal, Omniverse | Limited today |
| Geography | China, Japan, Taiwan, Singapore; no Vietnamese | Vietnam proving ground, Europe next |
Sources
- Product screens: datamesh.com product pages and FactVerse screenshots (AI Agent, FactVerse, predictive maintenance, Director, Simulator), captured 29-30 Oct 2026
- Docs: MCP tool reference, MCP guide; pricing
- News: industrial world models, Faurecia, Jebsee, 2024 funding
- Company: KDDI, EqualOcean B2, TNGlobal, LinkedIn
Conflicts left visible: customer counts (500+, 400+, 200+); Microsoft 2020 award "finalist" vs "winner"; MCP tools 21 at launch vs 103 in docs. Revenue and total funding after 2022 are not disclosed.