Products / In-process AI

jitCast

State-of-the-art forecasting inside your Java services, up to 13× faster.

Talk to us about jitCast jitCast
At a glance● data stays put
13×faster than the fastest PyTorch setup (TiRex-2, one series, A100)
<0.1%difference from the reference on the fev-bench and GIFT-Eval benchmarks
3leading models: Amazon Chronos-2, Mistral AI Timestral, NXAI TiRex-2
0training runs: forecasts new series on day one
The problem

Forecasts that arrive too late

New forecasting foundation models from Amazon, Mistral AI and NXAI predict demand, load, prices and risk without being trained on your data. But they ship as Python research code: slow to answer, and one more service to run beside the Java systems that act on the forecast.

What Beestream does

Forecasts inside the services that act on them.

jitCast runs Chronos-2, Timestral and TiRex-2 inside your Java services, on the GPU. It gives the same forecasts as each model's own engine, up to 13× faster, behind one API, so you can switch models as better ones appear.

What you get
  • No training project: zero-shot models forecast new series from their history alone
  • Planning ranges, not just a number: every forecast comes with quantiles
  • One API for three model families: change model without changing code
  • Biggest gains where speed matters most: real-time forecasts for single series and small batches
  • Runs where the decision is made, on NVIDIA or Apple GPUs, with no Python service to operate
Forecast speed-up over the fastest PyTorch setup, one series · higher is better
PyTorchfastest of eager, compile, TF32, BF16, FP16
1×
Chronos-2Amazon · jitCast
4.6×
Timestral-5MMistral AI · jitCast
10.7×
TiRex-2NXAI · jitCast
13.1×
NVIDIA A100 80GB · end-to-end forecast() · fastest configuration on each side
Where it pays off

Built for the decisions your business runs on.

Retail and supply chain

Demand forecasts per product and store, refreshed as sales come in.

Energy and utilities

Load and generation forecasts for grid balancing and trading.

Finance

Cash-flow, volume and risk forecasts inside existing Java systems.

IT and operations

Capacity planning and early warnings from metrics and telemetry.

The Beestream platform

Start with one product. Add the rest on the same GPUs.

jitCast is one part of the Beestream platform. Once it is in, the other products run on the same GPUs, under the same support, with no new stack to adopt.

7.3×Apache Iceberg GPU accelerationfaster Iceberg compaction UDFsSpark RAPIDS GPU accelerationcompiled to GPU kernels 19×JVector GPU accelerationfaster vector index builds 1.4×jitLLMllama.cpp prompt speed 2.4×jitNNfaster than PyTorch, ResNet-50
Contact

Tell us what is costing you CPU time. We'll show you the GPU path.

LinkedIn/company/beestream-ai X@beestream_ai Bluesky@beestream.ai GitHubgithub.com/beestream-ai