EmbeddingGemma.NET
A .NET binding for Google DeepMind's EmbeddingGemma-300m model that runs fully local, offline text embedding with no API key or cloud dependency, distributed as a single NuGet package for ASP.NET Core and Microsoft Semantic Kernel.
EmbeddingGemma.NET brings Google DeepMind's EmbeddingGemma-300m embedding model into .NET applications through ONNX Runtime, running entirely on-device so no embedding calls or data leave the machine. A single NuGet package registers the service either through the plain IServiceCollection used by ASP.NET Core and other generic .NET hosts, or through Microsoft Semantic Kernel's IKernelBuilder.
EmbeddingGemma is reported as the top-ranked open multilingual embedding model under 500 million parameters on the MTEB benchmark, supports over 100 languages, can run on low-end devices without a GPU with as little as 4 GB of RAM, and exposes 15 built-in task types that automatically apply the correct prompt prefix for a given use, such as retrieval or classification.
Before first use, the ONNX model and tokenizer files must be downloaded to disk, either with a bundled PowerShell script or by manually fetching five listed files, including a roughly 1.2 GB weights file, into a single directory.