CodeBrix.Ollama.ModelRunner.MitLicenseForever

A cross-platform, zero-dependency .NET library that runs a GGUF LLM model in-process and exposes it to the application through an interface contract: completion, streaming chat, embeddings, tool calling, grammar-constrained output, tokenization, model metadata and quantizing a model file to a smaller one. Binds a self-built native inference engine through hand-written P/Invoke, with native libraries for Windows x64 and ARM64, macOS Intel and Apple Silicon, and Linux x64, ARM64 and RISC-V 64. It also runs ONNX models on a pure-managed interpreter inside the package - no ONNX runtime, no Python, no native library and no NuGet dependency, so a graph runs wherever .NET runs: tensors in by name and tensors out by name for a graph of your own, MIDI music generated from a MIDI model and streamed event by event as it is written, and text generated from an ONNX bundle through the same interface contract, with a managed byte-level byte-pair tokenizer read from the bundle's own files. No server and no external Ollama installation; pair with CodeBrix.Ollama.ModelManager to resolve model names to files, and install both packages at the same version.