The Kelvin Platform
Autonomous control for industrial operations
Kelvin is a platform for building and deploying autonomous control applications on industrial equipment. From a pip install to a running SmartApp in under 20 minutes.
How it works
The closed-loop control cycle
Kelvin SmartApps operate in a continuous loop: read from the asset, compute a recommendation, act, and record the outcome. No manual intervention required between cycles.
Detect
SmartApps stream data continuously from connected assets using rolling windows. Events and anomalies are identified in real time as equipment operates.
Recommend
Based on incoming data, the SmartApp generates a control recommendation and publishes it to the connected asset. Each recommendation carries a full audit trail.
Prove
Every action and outcome is logged. Engineers review telemetry, approve or reject recommendations, and the system captures that signal to improve over time.
Safe autonomous control
Governed agents, not black boxes
Autonomous control in industrial environments requires more than accuracy. It requires a system that can be audited, constrained, and overridden. Kelvin is built with those requirements as first-class concerns.
Real-time guardrails
Every control recommendation is evaluated against configurable limits before it reaches the asset. Recommendations outside defined operating bounds are blocked automatically.
Audit trail on every action
Kelvin logs each recommendation, the data that produced it, and the outcome. Engineers can review the full decision history at any time.
98% engineer approval rate
Across production deployments, engineers approve more than 98% of recommendations generated by Kelvin SmartApps, reflecting trust built through transparency.
SOC 2 compliant
The platform meets SOC 2 security and availability standards, with offline-capable edge deployment and secure data syncing back to the cloud.
Data connectivity
Built for how industrial data actually moves
Kelvin connects to assets through the protocols already used in industrial environments. Data streams are ingested at the edge, available to SmartApps as rolling time windows, and synced back to the cloud with configurable retention.
Asset and datastream definitions can be imported in bulk via CSV or API, and visualized through the built-in data explorer or a Grafana integration.
Supported protocols
OPC-UA
Plant and equipment data
ModBus
Industrial device signals
MQTT
Streaming operational data
OSI PI
Historian integration
ROC
Oilfield controller data
SQL
Operational databases
Export and monitoring
- CSV and API-based data export
- Grafana integration for telemetry monitoring
- App logs and telemetry storage for troubleshooting
- Data annotation to improve SmartApp performance
Core capabilities
Three things the platform does well
Python SDK and app templates
Build SmartApps with the Kelvin Python SDK. Pre-built templates cover multi-objective optimization, event detection, Scikit-learn inference, and TensorFlow image recognition. Install with pip and deploy in minutes.
ML model integration
Import trained models from Databricks MLflow and AWS SageMaker directly into a SmartApp. The platform handles model serving at the edge without requiring changes to how models are built.
Edge deployment and scale
SmartApps run on commercial off-the-shelf hardware close to the equipment they control. Kelvin provisions a lightweight Kubernetes cluster automatically and scales to hundreds or thousands of edge devices.
Technical approach
Open architecture, runs anywhere
Any infrastructure
Deploy on any cloud provider, on-premises servers, or commercial off-the-shelf edge hardware. No proprietary infrastructure required.
Kubernetes-native
Kelvin provisions a lightweight K3S Kubernetes cluster automatically, or integrates with an existing Kubernetes environment.
Offline operation
Edge nodes run continuously without a cloud connection. Control continues during network outages, with secure data syncing when connectivity is restored.
API-first design
Every capability is accessible via API. Asset definitions, datastream imports, and telemetry export are all API-driven.
AI or first-principles
SmartApps can be built with ML models or with deterministic first-principles logic. Both types run on the same platform and follow the same deployment path.
SOC 2 compliant
The platform meets SOC 2 Type II security and availability standards across its cloud and edge components.
See Kelvin running on real equipment
Get a walkthrough of SmartApp deployment, from SDK install to live control recommendations on an industrial asset.
Questions? Email info@kelvin.ai