Skip to content
Docs
Open source

Documentation

Powerful AI.
Closer to home.

Run open models on the Intel hardware you already own. From your first local request to a multi-machine inference pipeline.

  • Open source
  • OpenAI-compatible
  • Built for Intel
Check compatibility

01 / Get started

Your first step starts here. One machine or many, same starting point.

Download Cascadia, serve a ready-made model, and send it a request. No build required.

Put your Intel hardware to work. Export a model and serve it on one machine.

02 / Capabilities

A small footprint. A bigger possibility.

Multi-machine inference

Run models larger than a single machine. Connect your Intel devices into a pipeline.

OpenAI-compatible API

Keep your favorite tools and SDKs. Point your application at a local endpoint.

Built-in model sharding

Export and quantize Hugging Face models into ready-to-run INT4 stages.

Intel-native engines

Choose the right engine for your model, from OpenVINO to specialized MoE paths.

Visibility at every stage

Track token throughput, latency, and pipeline health with Prometheus metrics.

Inference on your terms

Keep inference on your own network. Understand the deployment security model.

03 / Reference

Go a level deeper. The commands, internals, and interfaces behind Cascadia.