Ulysses is an Energy Based Model for streaming time series and image inference — active learning, real-time, on the CPUs you already have.
Continuous monitoring of streaming data with sub-10ns detection latency. Identify deviations from learned patterns without explicit anomaly labels.
Process high-frequency time-series data in continuous streams. No batching, no windowing artifacts. Native support for multi-dimensional sensor feeds.
Full inference capability on edge hardware without cloud dependency. Designed for disconnected, intermittent, and low-bandwidth environments.
Direct memory access inference pipeline eliminates data serialization overhead. No copying between CPU and accelerator — because there is no accelerator.
Optimized for CPU. Zero GPU dependency means deployment on existing infrastructure with no procurement cycle and no specialized hardware.
Sits in front of LLMs, ML pipelines, and domain-specific models as a sensory pre-processor. Enriches downstream systems with real-time physical-world context.