Ekkono

paid

Ekkono enables ML training and inference directly on constrained edge devices. Incremental learning, federated learning, and tiny footprint for sensor-data use cases.

Data & Analytics

AI Models & Infrastructure

AI Frameworks

About

Ekkono is a commercial Edge AI platform designed for embedded systems engineers, data scientists, and application developers who need to deploy smart, self-learning capabilities directly onto resource-constrained hardware. Unlike TensorFlow Lite/Micro, which is optimized for deep learning tasks such as image classification, Ekkono is built specifically for real-world edge ML use cases involving sensor data — temperature, pressure, power consumption, and more. At the heart of Ekkono is its unique incremental learning capability, allowing ML models to train continuously on each individual device rather than relying solely on cloud-trained models. This enables genuine per-unit condition monitoring, adaptive automation, and real-time anomaly detection tailored to how each individual unit actually behaves in the field. The platform ships as C and C++ libraries with no external dependencies, making it hardware and sensor agnostic — deployable on everything from an Arduino Pro Micro with 2 KB of RAM to industrial controllers. Ekkono Synthesis handles model life-cycle management (MLCM) at scale, aggregating learnings across device fleets through federated learning, outlier detection, and model correlation. Key solutions include Condition Monitoring (virtual health-indicator sensors), Condition-Based Maintenance (remaining useful life estimation), Performance Optimization (real-time settings simulation), and Sustainable Operations (ensuring devices run only as needed). Ekkono is ideal for OEMs and industrial companies seeking to transition from reactive, rule-based automation to proactive, predictive, and personalized device intelligence.

Key Features

On-Device Incremental Learning: Trains ML models continuously on each edge device using real-time sensor data, enabling per-unit personalization without requiring a cloud round-trip.
Federated Learning via Ekkono Synthesis: Aggregates models across entire device fleets through model correlation, outlier detection, and federated learning to improve collective intelligence at scale.
Ultra-Tiny Footprint: C and C++ libraries with no external dependencies, capable of running on hardware as constrained as 2 KB RAM Arduino devices or 64 KB RAM Commodore 64-class chips.
Hardware & Sensor Agnostic: Runs on any CPU architecture with any sensor input — pressure, temperature, power, or custom streams — with hotswappable ML models and no vendor lock-in.
Full Model Life-Cycle Management (MLCM): Ekkono Synthesis handles deployment, versioning, aggregation, and fleet-wide model management to support production-grade edge ML operations.

Use Cases

Industrial equipment condition monitoring using virtual sensor models trained on each unit's individual normal behavior to detect deviations early.
Predictive maintenance for wear parts by estimating remaining useful life (RUL) from real-time edge sensor data, enabling proactive servicing schedules.
Adaptive performance optimization of machines and systems by continuously simulating and applying optimal settings based on real operating conditions.
Federated learning across large device fleets to aggregate model improvements from individual units without sharing raw sensor data.
Sustainable product operations by ensuring every unit runs efficiently and only when needed, reducing energy consumption and environmental footprint.

Pros

True Edge Training, Not Just Inference: Most edge AI platforms only run pre-trained models. Ekkono uniquely supports incremental training on-device, enabling adaptive, self-learning products.
Extremely Minimal Hardware Requirements: Designed to run on the most constrained embedded hardware, opening AI capabilities to product categories previously considered too limited for ML.
Commercial Support & Production Readiness: Unlike open-source alternatives, Ekkono provides dedicated commercial support and MLCM tooling suited for industrial production deployments.
Purpose-Built for Sensor Data: Optimized for the most common real-world edge ML scenarios — sensor-based monitoring and control — rather than image classification workloads.

Cons

Not Suited for Deep Learning / Computer Vision: Ekkono focuses on sensor data ML (shallow models), so teams needing image classification or NLP at the edge should look to TensorFlow Lite/Micro instead.
Enterprise Pricing Barrier: As a commercial B2B SDK, Ekkono is likely cost-prohibitive for indie developers or small startups without established hardware product lines.
Requires Embedded Engineering Expertise: Integration involves C/C++ development and embedded system knowledge, meaning teams without hardware engineering background face a steep learning curve.

Frequently Asked Questions

TensorFlow Lite/Micro is designed for deep learning inference tasks like image classification. Ekkono is purpose-built for sensor-data ML with unique capabilities like incremental on-device training, real-time anomaly detection, and federated learning — features TensorFlow Lite/Micro does not offer.

Ekkono runs on extremely constrained hardware with as little as 2 KB of RAM, such as an Arduino Pro Micro. Its C and C++ libraries have no external dependencies, making it compatible with virtually any CPU architecture.

Incremental learning allows an ML model to update itself continuously from new data without being fully retrained from scratch. At the edge, this means each device adapts to its own real-world operating conditions, enabling true per-unit personalization and adaptive automation.

Ekkono Synthesis is the fleet-level model life-cycle management (MLCM) layer. It aggregates what individual devices have learned collectively through federated learning, model correlation, and outlier detection — allowing a fleet of devices to improve together.

Ekkono is primarily used for condition monitoring (detecting anomalies and health indicators), condition-based maintenance (predicting remaining useful life of components), performance optimization (real-time tuning of device settings), and sustainable operations (ensuring devices run only when and as needed).