Asset risk
Transformer TR-2048
- Evidence
- 7 supporting sources
- Related assets
- 12
- Open work orders
- 2
- Policy check
- Passed
Recommended next actionSchedule engineering review
Human approvalHuman approval required
GridSight · by NeuroCluster
Turn grid data into controlled action.
GridSight connects operational, asset and enterprise data through a governed grid ontology — giving AI the context, authority and evidence required to support critical grid decisions.
Not a dashboard. Not another data lake. The layer that makes the systems already running your grid answer one question together.
Built for customer-controlled, sovereign and regulated environments.
Physical network
Operational & enterprise sources
- SCADA
- GIS
- SAP
- Weather
- Inspection
- Work orders
- Market data
GRIDSIGHT
Grid ontology · Governed agents · Evidence
NeuroCluster Intelligence OS
- Context
- Risk
- Forecast
- Decision
- Action
- Runs inside your environment
- Evidence on every answer
- Policy before action
- Human approval where it matters
The problem
Europe's grid doesn't have a data problem. It has a context problem.
Grid operators already generate enormous volumes of operational and enterprise data. The information required for a single decision is spread across systems that were never designed to answer a question together.
Every operator has solved the transport problem. Data flows between systems, warehouses fill, dashboards render. What no pipeline delivers is the relationship between a transformer, the line it feeds, the defect an inspector photographed last spring, the work order already raised against it and the operational policy that decides who may act.
That relationship is reassembled by hand, by an engineer, on every single decision. It is the slowest part of the work and the first thing lost when the person who held it in their head moves on.
- SCADAWhat is happening now?
- GISWhere is the asset?
- SAP / EAMWhat happened before?
- InspectionsWhat condition is it in?
- WeatherWhat happens next?
- Work ordersWhat are we already doing?
- Market dataWhat external conditions matter?
Critical decisions require all of them.
Traditional integration
Traditional integration moves data.
GridSight
GridSight connects its meaning.
The magic moment
From scattered signals to one operational picture.
One question, asked two ways.
Which assets require attention before they become operational risks?
The same sources, resolved against one asset through the grid ontology.
- Asset identity
- Physical relationships
- Current operating state
- Historical maintenance
- Inspection evidence
- Environmental conditions
- Existing work
- Operational policies
One governed recommendation.
Elevated attention recommended
Evidence
Every claim resolves to the record it came from.
Policy
The recommendation is checked against your rules before it is shown.
Human control
A qualified person decides. GridSight prepares the case.
Illustrative example. Asset identifiers and counts are shown to demonstrate the output format, not results from a deployment.
How it works
From grid data to controlled action.
Five stages. Each one is inspectable, each one is governed, and none of them require you to move your systems of record.
Operational, spatial, enterprise and external sources are connected where they live. Operational technology is read through a one-directional, rate-limited broker — typically a historian rather than the control system itself.
- SCADA
- GIS
- SAP
- EMS
- EAM
- IoT
- Weather
- Documents
- Inspection
- Market data
Every answer traces back to evidence.
Every action remains subject to enterprise policy.
GridSight does not issue control commands to process systems. Anything that changes the state of the network goes to a qualified operator, inside the systems and procedures already approved for that purpose.
The grid ontology
Give AI a model of the grid — not just access to its data.
Retrieval finds documents that mention a transformer. It cannot tell you which line that transformer feeds, or which other assets share its defect.
A language model can retrieve documents.
GridSight understands relationships.
Search and retrieval treat your enterprise as a pile of text. That works for questions whose answer is written down somewhere. It fails for the questions grid operators actually ask, because those answers are not written anywhere — they exist in the relationship between records held in different systems.
The ontology makes those relationships first-class. An asset has a position in the topology, a condition history, an inspection record, a risk, an owner and a policy. Reasoning becomes traversal instead of guesswork.
- SubstationcontainsTransformer
- Transformerconnected_toLine
- LinesuppliesRegion
- TransformerhasInspection
- InspectionidentifiesDefect
- DefectcreatesRisk
- Riskmitigated_byWork order
- Work ordergoverned_byPolicy
- PolicyrequiresApproval
The ontology becomes reusable operational context.
The first use case builds context. The second reuses it.
Every subsequent workflow becomes cheaper to deploy, because the enterprise model already exists. This is the difference between an AI project and an intelligence layer: the second one does not start from zero.
- 1Use case 01Builds the objects, relationships and connections it needs.
- 2Use case 02Reuses most of that model and extends it at the edges.
- 3Use case 03+Mostly configuration — the semantics are already there.
Applications
One intelligence layer. Multiple grid applications.
Three application areas over the same ontology, the same identity model and the same governance. Adding the second does not mean rebuilding the first.
GridSight Operations
Operational intelligence across grid state, events, assets and decisions.
Bring the current state of the network together with the context needed to interpret it, so an operator investigating an event is not reconstructing it from four screens.
- Operational situation awareness
- Event correlation
- Topology-aware investigation
- Operator decision support
- Evidence-backed incident analysis
GridSight Asset Intelligence
Connect asset history, inspection, maintenance and operating conditions.
Treat condition as something assembled from every source that observes it — telemetry, inspection imagery, maintenance records and operating environment — rather than a field in one system.
- Asset condition intelligence
- Inspection analysis
- Anomaly correlation
- Maintenance prioritisation
- Work-order intelligence
- Evidence-based asset risk
GridSight Contingency
Contextual intelligence for network risk and contingency workflows.
Support the analysis around contingency work: what an asset relates to, what a scenario changes, which constraints bear on it, and what the evidence says.
- N-1 analysis support
- Asset relationship reasoning
- Scenario comparison
- Operational constraints
- Evidence-backed recommendations
- Human approval workflows
These are decision-support applications. GridSight does not independently perform protection, switching or safety-critical control, and does not replace the certified systems that do.
Asset in context
Select an asset. See everything that bears on it.
A simplified network, from generation through transmission to critical load. Choose any asset to see the operational picture GridSight assembles for it.
NominalWatchAttention
Illustrative model with example data. It demonstrates the shape of the output — asset identifiers, readings and counts are not from a live deployment.
Agents
Agents that understand the enterprise before they act.
A GridSight agent is not a chat window with grid documents behind it. It carries an identity, a mandate and a set of permissions, and it plans against the ontology.
Question
“Which substations have elevated operational risk this week?”
Agent plan
- 01Find relevant assets
- 02Query telemetry
- 03Check inspection evidence
- 04Retrieve maintenance history
- 05Check work orders
- 06Evaluate weather exposure
- 07Apply risk policy
- 08Generate recommendation
Control
- Identity verified
- Permissions checked
- Policy passed
- Evidence attached
- Human approval required
Result
Recommendation, evidence and a next action — inside the workflow that owns it.
Intelligence without authority is a chatbot.
Intelligence with uncontrolled authority is a risk.
GridSight governs both.
Human in the loop
Autonomy where appropriate. Human control where required.
Autonomy is not a product setting with two positions. It is a spectrum, configured per workflow and enforced by the control plane.
01
Observe
AI reads and correlates information.
Human: Informed
02
Recommend
AI proposes an action with its evidence.
Human: Decides
03
Prepare
AI assembles the workflow or transaction.
Human: Reviews
04
Approve
A qualified person validates the proposal.
Human: Authorises
05
Execute
An approved system performs the permitted action.
Human: Accountable
Configurable per
- Workflow
- Asset class
- System
- User
- Agent
- Risk level
- Policy
The level that applies is a property of the workflow, not of the model. Changing model does not change what an agent is allowed to do.
Deployment
Deploy inside the environment you already control.
GridSight runs where your policies say it can run. Adopting it does not require moving data to a vendor platform or buying infrastructure from us.
Customer datacenter
Your hardware, your network, your operational control.
Private cloud
Your tenancy in the cloud you already govern.
Sovereign European cloud
Deployment in an EU-jurisdiction environment of your choosing.
Edge / OT environment
Local processing close to distributed assets.
Air-gapped
Disconnected operation with local inference.
Hybrid
Split by workload, data class or jurisdiction.
Your infrastructure. Your models. Your data. Your policies.
NeuroCluster is model-independent and infrastructure-independent. GridSight can run against customer-selected models, European models, open-weight models, private models or specialised domain models.
Owning NeuroCluster infrastructure is an option, not a prerequisite. The platform is designed to be deployed into an environment the customer already controls.
Security & governance
Built for environments where “probably safe” is not enough.
Governance is not a settings page bolted onto a model. It is the layer every request passes through before it reaches data, a tool or a workflow.
Identity & access
Every user and every agent authenticates and carries a defined scope.
Agent authority
An agent has an owner, a mandate and permissions it cannot exceed.
Policy enforcement
Rules are evaluated before an answer is produced, not audited afterwards.
Data isolation
Tenancy, classification and residency boundaries are enforced in the runtime.
Model routing controls
Which model may see which class of data is a policy decision.
Evidence lineage
Every claim retains the source it was derived from.
Audit logging
Requests, tool calls, approvals and outcomes are recorded.
Human approval
Approval gates are part of the workflow definition.
Deployment isolation
Workloads are separated by environment and by sensitivity.
Observability
Model, agent and workflow behaviour is traceable in operation.
Designed to support organisations operating under
GDPR
Data-protection obligations over personal data in operational and enterprise records.
EU AI Act
Governance expectations for AI systems used in regulated and high-impact settings.
NIS2
Cyber-resilience and incident obligations for essential-entity operators.
CRA
Product security expectations across the software supply chain.
These are architectural capabilities intended to support your compliance programme. NeuroCluster does not claim certification or formal conformity assessment against these frameworks on your behalf.
Model independence
The best model today should not become tomorrow's lock-in.
Different workloads need different models. Classification, long-context reasoning, vision over inspection imagery and local edge inference do not share one optimum.
- Mistral
- Llama
- Qwen
- Customer models
- Specialised models
- Future models
NeuroCluster model router
Routes by workload class, data classification and jurisdiction
GridSight
Performance
Match the model to the difficulty of the workload.
Cost
Route routine work to cheaper models and reserve capacity for hard problems.
Jurisdiction
Keep classes of data inside an approved legal boundary.
Security
Restrict sensitive context to models running under your control.
Latency
Serve interactive and edge workloads locally.
Specialisation
Use domain or vision models where general models underperform.
Models change.
Enterprise context persists.
GridSight separates enterprise intelligence from the model provider. Replacing a model is a routing change, not a rebuild of your ontology, policies, agents or evidence trail.
Integration
Keep the systems that already run the grid.
GridSight complements the operational and enterprise landscape. Your systems of record stay your systems of record.
Operations
- SCADA
- EMS
- DMS
- Historian
Read through a one-directional broker.
Enterprise
- SAP
- ERP
- EAM
- CRM
Remain the system of record.
Spatial
- GIS
- Asset maps
- Network topology
Anchors assets in the network model.
Data
- Databases
- Data lakes
- Warehouses
- Streams
Connected where they already live.
Unstructured
- Documents
- Reports
- Manuals
Typed against objects, not filed as text.
External
- Weather
- Market information
- Regulatory data
- Geospatial sources
Context the grid does not generate itself.
System categories describe intended integration coverage. They do not imply partnership, certification or endorsement by the named vendors.
Adoption
Start with one decision. Build an intelligence layer.
The failure mode in enterprise AI is modelling everything before delivering anything. GridSight is designed to be adopted the other way around.
- 01
Land
One governed workflow, with the objects and connections it actually needs.
A six to ten week deployment is the target where the workflow, data access and approvals are ready.
- 02
Prove
Demonstrate operational value on that workflow with evidence attached.
- 03
Expand
Add assets, workflows and data sources against the same model.
- 04
Standardise
Shared ontology and governance become reusable infrastructure.
- 05
Scale
Agents and applications operate across business units.
Every deployment compounds the value of the shared context.
The six to ten week figure is a planning target, not a contractual commitment. Actual duration depends on data access, system readiness and internal approvals.
Why NeuroCluster
Not another layer of lock-in.
Six primitives. Every GridSight capability is built from them, and so is every other application on the platform.
- 01
Context
A living ontology connects enterprise meaning across systems.
- 02
Identity
Every user and agent operates with defined authority.
- 03
Policy
Actions remain inside organisational rules.
- 04
Intelligence
Use the right model for each workload.
- 05
Evidence
Decisions remain traceable to their sources.
- 06
Action
Intelligence connects back into operational workflows.
ContextIdentityPolicyIntelligenceEvidenceAction
Sovereignty
Sovereignty is operational freedom.
Server location is the easiest part of sovereignty and the least useful on its own. What matters is whether you can still change your mind.
Model freedom
Change models without rebuilding the enterprise layer.
Infrastructure freedom
Change deployment environment without changing the application.
Data control
Keep sensitive data inside a boundary you define.
Identity control
Retain authority over which agents and users can do what.
Auditability
Know why a decision was made, after the fact.
Exit freedom
Avoid permanent dependence on a single AI vendor.
Sovereign by architecture, not by marketing.
Outcomes
Built around operational outcomes.
We do not publish savings percentages for deployments that have not happened. These are the categories of change GridSight is built to produce.
01
Faster investigation
Bring fragmented evidence into one operational context.
02
Better asset decisions
Connect condition, history, topology and operational state.
03
Controlled AI adoption
Introduce agents without surrendering governance.
04
Reusable context
Stop rebuilding integrations and semantics for every AI project.
05
Model optionality
Adopt better models without rebuilding the enterprise layer.
06
Auditable decisions
Preserve evidence and decision history.
For architects
Under the hood.
Optional depth. Expand what is relevant to your review.
Context & reasoning
Ontology
Typed objects, named relationships, events and rules over the grid domain. Objects resolve across source systems rather than being copied into a new master.
Graph relationships
Traversal over the object model answers impact and dependency questions without a bespoke query per question.
Vector retrieval
Semantic retrieval over documents and findings, typed against the objects they describe rather than returned as loose text.
Memory
Scoped, inspectable state across a task or an investigation, subject to the same permissions as the underlying data.
Agents & models
Agent orchestration
Agents carry an owner, a mandate, scoped permissions and an evidence trail. Plans are produced before tools are called.
MCP
Tools and data sources are exposed to agents through a declared interface, so an agent's reach is a configuration decision.
Model routing
Workload class, data classification and jurisdiction determine which model serves a request.
Private inference
Open-weight and private models served inside the customer boundary for sensitive workloads.
Control & runtime
Identity
Enterprise identity for users and first-class identity for agents, with authority defined per workflow.
Policy engine
Externalised policy evaluated at request time, covering data access, tool use and action authority.
Evidence lineage
Retained provenance from answer to source record, so a decision can be reconstructed.
Observability
Tracing across model calls, tool calls and workflow steps for operational and review purposes.
Kubernetes
The runtime is deployed as a Kubernetes workload, which is what makes the deployment options interchangeable.
Edge & air-gapped
Local processing at distributed assets and fully disconnected operation with local inference.
Platform
GridSight is where NeuroCluster starts. The Intelligence OS is where it expands.
Energy is a demanding proving ground: long-lived physical assets, real consequence, strict governance and systems that cannot be disturbed. The primitives that survive it generalise.
Applications
GridSightEnergy & gridAvailableIndustrial intelligenceManufacturing & industryDirectionFinancial intelligenceFinancial infrastructureDirectionGovernment intelligencePublic sectorDirectionEnterprise operationsCross-industryDirection
NeuroCluster Intelligence OS
The same primitives power every domain
- Ontology
- Identity
- Policy
- Agents
- Evidence
- Models
- Workflows
- Deployment
Adjacent areas describe where the platform architecture is designed to extend. They are directions for the platform, not shipping products with their own release schedule.
Questions
The questions that decide procurement.
- Does GridSight replace our SCADA, GIS or asset management systems?
- No. GridSight is an intelligence layer over the systems that already run the grid. Your asset management system stays the system of record; the ontology defines what its objects mean in relation to topology, condition, work and obligations held elsewhere.
- Does anything connect directly to our control systems?
- Operational technology is read through a one-directional, rate-limited broker, and usually via a historian rather than the control system itself. There is no write path from GridSight into process control.
- Can GridSight perform switching or grid control?
- No. GridSight is decision support. It assembles the case, checks it against policy and proposes an action. Anything that changes the state of the network goes to a qualified operator inside the systems and procedures already approved for that purpose.
- Can it run without sending data outside our environment?
- Yes. GridSight can be deployed in a customer datacenter, a private or sovereign cloud, at the edge, or fully air-gapped with local inference on open-weight or private models.
- How long before an ontology is useful?
- Useful comes from modelling one decision rather than the whole estate — typically the asset, its topology, its work history and the constraint that bears on it. Enterprise-wide modelling before a first workload reliably produces something nobody uses.
- What happens when we want to change the underlying model?
- Model selection is a routing decision. The ontology, policies, agent mandates and evidence trail are independent of the model provider, so replacing a model does not require rebuilding the enterprise layer.
Start with the decision that matters most.
- 01Choose one high-value grid workflow.
- 02Connect the required systems.
- 03Build the operational context.
- 04Deploy governed intelligence inside your environment.
Request a GridSight demo
Tell us the decision you want to improve. We will shape the session around it.