Time Series.
For the Future.

We deliver real-time time series predictions in seconds.
Up to 100 times faster than conventional models.

Thank you! Your submission has been received!

Oops! Something went wrong while submitting the form.

Lightning-Fast.
Predictions in under five seconds, up to 100× faster than conventional methods.

Edge Ready.
Runs seamlessly on laptops, microcontrollers, and industrial edge devices.

Next-Gen Precision.
Proven accuracy in critical energy and industrial applications.

Figure adapted from Yan, M., Huang, C., Bienstman, P. et al. Emerging opportunities and challenges for the future of reservoir computing. Nat Commun 15, 2056 (2024). https://doi.org/10.1038/s41467-024-45187-1.

Licensed under CC BY 4.0 http://creativecommons.org/licenses/by/4.0/

Reservoir Computing

Reservoir Computing (RC) offers compact models that can run on devices with limited memory—often below one megabyte—yet scales to billions of parameters if needed.

Deep leaning (DL) models demand terabytes of storage and specialized hardware. A small RC network, with just a few thousand neurons, can execute efficiently on a typical microcontroller or smart grid monitor with a few megabytes of RAM.

Reservoir computing provides flexibility, speed, and power for robust, real-time forecasting across an extraordinarily diverse range of platforms

Reservoir Computing for Chaotic Time Series

Traditional ML Models

Entrox Reservoir Computing

Needs GPUs or server clusters for adequate speed

Runs on laptops or microcontrollers, reduces hardware costs up to 10-20x

Needs extensive labeled datasets for reliable training

Requires up to 1000x fewer samples to achieve similar accuracy (lightweight readout training)

Complex models can take minutes or hours for each forecast

Predictions in seconds, often 10-100x faster for chaotic, high-dimensional tasks

Struggles with non-linear, high-dimensional chaotic systems without specialized architectures

Designed to capture non-linear dynamics, shows excellent performance on benchmarks like Lorenz-96

High power costs from GPU/CPU usage

Uses much less power usage due to simpler training overhead

Time -consuming retraining cycles

Fine-tuning readout layer is straightforward, so updates require minimal effort

Complex to scale; often needs repeated training on large infrastructure

Low memory demands and minimal recalibration allow for simplified scaling.

The best on the market.

up to 100x
faster.

Than traditional machine learning models.

up to 1000x
cheaper.

Training costs scale with
computational demand.

Edge
ready.

Easily deploy your model on laptops,
microcontrollers, and older hardware.

Meet the team

Dr. Tamon Nakano

AI R&D, Mechanical Engineering, Business Development

Sebastian Baur

AI R&D, Software Development

Daniel Köglmayr

AI R&D, Algorithm Development

Tristan Pelser

Energy Sector, Marketing, Business Development

What we do

We empower businesses with reservoir computing, enabling ultra-fast, high-accuracy time series forecasting. Our technology accelerates chaotic system simulations—saving time and costs in energy optimization, smart meters, IoT, vibration analysis, and predictive maintenance.

Running efficiently on standard laptops and edge devices, our algorithm cuts setup time, reduces cloud dependency, and enhances data security. With unmatched speed, resource savings, and precision, we transform forecasting into a powerful competitive advantage.

Our Vision

Pushing Boundaries:
‍We aim to redefine how organizations predict and manage dynamic processes—delivering unmatched speed and accuracy in the most demanding scenarios.
‍
Democratizing AI:
By leveraging on-device computing, we make advanced analytics accessible and affordable for businesses of all sizes.
‍
Real-World Impact:
Our mission is to streamline adoption of cutting-edge predictive methods, boosting operational efficiency and sustainability across industries.

Time Series.For the Future.