InferenceEngine.Core

A cross-platform .NET 10 library for AI inference with ONNX Runtime — hardware acceleration, automatic fallback, and built-in diagnostics for provider selection.

InferenceEngine.Core wraps ONNX Runtime behind three small API surfaces (one-shot, streaming, and out-of-process), probes the host for the best execution provider (MIGraphX on AMD ROCm, DirectML on Windows, CoreML on macOS, CPU everywhere), and degrades gracefully when a provider's native stack is absent.

Install

dotnet add package InferenceEngine.Core

Prerelease (-dev / -rc) builds are published to GitHub Packages — see the README for the feed setup.

Guides

  • Consumer integration guide — start here: which of the three API surfaces fits your app, and how to wire each one up.
  • Streaming inference guide — chunked/token-by-token results for LLM/SLM runners, speech engines, and step-wise diffusion.
  • Text engines guide — string-in/string-out embedding and generation engines with tokenization owned by the library.

API reference

Full generated API documentation: API reference.

Coverage

Live line/branch coverage from main, rendered by ReportGenerator: coverage report.