OpenTail.Stingray 1.0.5

Prefix Reserved
There is a newer version of this package available.
See the version list below for details.
dotnet add package OpenTail.Stingray --version 1.0.5
                    
NuGet\Install-Package OpenTail.Stingray -Version 1.0.5
                    
This command is intended to be used within the Package Manager Console in Visual Studio, as it uses the NuGet module's version of Install-Package.
<PackageReference Include="OpenTail.Stingray" Version="1.0.5" />
                    
For projects that support PackageReference, copy this XML node into the project file to reference the package.
<PackageVersion Include="OpenTail.Stingray" Version="1.0.5" />
                    
Directory.Packages.props
<PackageReference Include="OpenTail.Stingray" />
                    
Project file
For projects that support Central Package Management (CPM), copy this XML node into the solution Directory.Packages.props file to version the package.
paket add OpenTail.Stingray --version 1.0.5
                    
#r "nuget: OpenTail.Stingray, 1.0.5"
                    
#r directive can be used in F# Interactive and Polyglot Notebooks. Copy this into the interactive tool or source code of the script to reference the package.
#:package OpenTail.Stingray@1.0.5
                    
#:package directive can be used in C# file-based apps starting in .NET 10 preview 4. Copy this into a .cs file before any lines of code to reference the package.
#addin nuget:?package=OpenTail.Stingray&version=1.0.5
                    
Install as a Cake Addin
#tool nuget:?package=OpenTail.Stingray&version=1.0.5
                    
Install as a Cake Tool

OpenTail.Stingray

The Unified Native Multimodal AI Engine for .NET 10
100% pure managed C# — Zero Python, Zero P/Invoke to llama.cpp, Zero external server sidecars.
Runs in-process on CPU (AVX-512/AVX2/NEON), Vulkan, or CUDA. Publishes as a single NativeAOT binary.

License: MIT .NET 10 NativeAOT NuGet

Built by opentail.net


Why Stingray?

In Python and C++, running modern local AI requires juggling 4–5 fragmented, heavy tools (llama.cpp for text, whisper.cpp for speech, kokoro for voice synthesis, and ComfyUI or diffusers for image/video generation).

Stingray unifies the entire local AI ecosystem into one clean, lightweight .NET library. All tensor math, SIMD kernels, KV caches, diffusion flow-schedulers, audio DSP resamplers, and tokenizers are written in high-performance managed C# that you can step into, profile, and deploy anywhere.


Superpowers at a Glance

  • 💬 Text LLMs & Sparse MoE: Llama 3 / 3.1 / 3.2 / 3.3 / 4, Qwen 2.5 / Qwen 3.5 MoE, DeepSeek-V3 / R1, Gemma 3 / 4, SmolLM2. Full tool calling, JSON schema constrained decoding, and grammar masks.
  • 👁️ Multimodal Vision: Native Gemma 4 unified (gemma4uv) & ViT (gemma4v), Gemma 3 SigLIP (gemma3), Llama 4 ViT, and GLM-4V patch grids.
  • 🎙️ Studio Audio Stack:
    • Speech-to-Text (STT): Whisper with $O(N)$ KV Cache, cross-attention projection, and Silero VAD.
    • 5 Neural Voice Engines (TTS): Kokoro, Piper (VITS), MeloTTS, Chatterbox, and F5-TTS Flow-Matching DiT with real-time clause streaming.
    • Broadcast DSP: Rational windowed-sinc resampler, ATSC A/85 downmixing, and TPDF dithered 16-bit/24-bit WAV exporter.
  • 🎨 State-of-the-Art Diffusion & Video:
    • Image Architectures: SD 1.5, SDXL, SD 3 / 3.5 (MMDiT), FLUX.1 (schnell/dev), FLUX.2 (Klein & Kontext multi-reference), FLUX 3 (3D/4D RoPE Multimodal), and Z-Image-Turbo.
    • Acoustic Diffusion: Stable Audio 3 (Variable-Length 44.1kHz Stereo DiT).
    • Video Diffusion: Wan 2.1/2.2 Video, HunyuanVideo, and LTX-Video.
    • Real-Time Streaming: Latent Consistency Models (LCM 1–4 step) and StreamBatchPipeline (30–60 FPS real-time webcam/video streaming).
    • Exporters: Animated GIF and animated PNG sequence writers with Floyd-Steinberg dithering.
  • ⚡ Zero-Latency Hardware Engine: Nanosecond SIMD evaluation (AVX-512, AVX2, ARM Neon), sub-millisecond cached GPU topology (~/.stingray/hardware_cache.json), and SmartOffloadPlanner for discrete GPUs, APUs (AMD Vega/Intel Iris), and CPUs.

Installation

dotnet add package OpenTail.Stingray

Quick Starts

1. LLM Chat with Streaming

using OpenTail.Stingray.Core;
using OpenTail.Stingray.Cpu;
using OpenTail.Stingray.Engine;

// 1. Open GGUF model and initialize hardware backend
var model     = GgufModel.Open("models/Llama-3.2-3B-Instruct-Q4_K_M.gguf");
var hp        = ModelHyperparams.FromGgufMetadata(model.Metadata, model);
var tokenizer = GgufTokenizer.FromGgufModel(model);
var backend   = new CpuBackend();
var forward   = new ForwardPass(model, backend, hp);

using var engine = new InferenceEngine(forward, tokenizer, modelId: "llama3", owned: [backend, model]);

var sampling = new SamplingParams
{
    Temperature  = 0.7f,
    TopP         = 0.9f,
    MaxNewTokens = 512,
    StopTokenIds = [.. tokenizer.EogTokenIds]
};

// 2. Stream generation in real-time
await foreach (var chunk in engine.GenerateChunksAsync("Explain quantum computing in simple terms:", sampling))
{
    if (chunk.Kind == GenerateChunkKind.Text)
        Console.Write(chunk.Text);
}

2. Studio Voice Synthesis (Text-to-Speech)

using OpenTail.Stingray.Audio;
using OpenTail.Stingray.Audio.Kokoro;

ITextToSpeechPipeline tts = new KokoroPipeline();
await tts.LoadAsync("models/kokoro-v1.0.safetensors");

// Synthesize high-fidelity 24kHz audio
AudioBuffer audio = await tts.SynthesizeAsync("Welcome to OpenTail Stingray.", voice: "af_heart", speed: 1.0f);

// Save with broadcast-quality TPDF dithering
WavWriter.Write("output.wav", audio);

3. Image Generation with FLUX.2 or SDXL

using OpenTail.Stingray.Diffusion;
using OpenTail.Stingray.Diffusion.Flux2;

var pipeline = new Flux2Pipeline();
await pipeline.LoadAsync("models/flux-2-klein-4step.safetensors");

var prompt = new DiffusionPrompt
{
    Positive = "A cinematic shot of an astronaut riding a horse on Mars, 8k, photorealistic",
    Width = 1024,
    Height = 1024,
    Steps = 4,
    GuidanceScale = 3.5f
};

var rgb = await pipeline.GenerateAsync(prompt);
PngWriter.Write("astronaut.png", rgb, prompt.Width, prompt.Height);

What's in the Package

One dotnet add package OpenTail.Stingray brings the complete, unified engine:

Assembly Modality & Functionality
OpenTail.Stingray.Core GGUF/Safetensors parser, BPE/SentencePiece tokenizers, chat templates, grammar masks, HardwareCapabilities zero-latency profiler, and SmartOffloadPlanner.
OpenTail.Stingray.Engine Autoregressive forward pass, paged KV cache, speculative decoding, continuous batching, and sampling.
OpenTail.Stingray.Cpu CPU SIMD kernels (AVX-512/AVX2/NEON), fast dequantization (Q4_K, Q8_0), and multithreaded GEMM.
OpenTail.Stingray.Vulkan High-performance Vulkan compute backend with precompiled SPIR-V shaders.
OpenTail.Stingray.Cuda CUDA backend with cuBLAS GEMM and runtime NVRTC shader compilation.
OpenTail.Stingray.Vision Unified vision projection for Gemma 4 (gemma4uv/gemma4v), Gemma 3 SigLIP, Llama 4, and GLM-4V.
OpenTail.Stingray.Audio Studio Audio: Whisper STT ($O(N)$ KV Cache + Silero VAD), 5 Neural TTS engines (Kokoro, Piper, Melo, Chatterbox, F5-TTS), rational sinc resampler, ATSC A/85 downmixing, and TPDF dithered WAV export.
OpenTail.Stingray.Diffusion Diffusion & Video: SD 1.5, SDXL, SD 3/3.5, FLUX.1, FLUX.2, FLUX 3 Multimodal, Stable Audio 3, Z-Image-Turbo, Wan 2.1/2.2, HunyuanVideo, LTX-Video, LCM 1–4 step distillation, StreamDiffusion pipelining, and GIF/PNG sequence exporters.
OpenTail.Stingray.Pipeline Three-tier VRAM → RAM → NVMe weight streaming for MoE expert offload.
OpenTail.Stingray.TurboQuant KV-cache compression (KVarN 4/2-bit, Lloyd-Max 3/4-bit).

  • OpenTail.Stingray.Server: Drop-in ASP.NET Core endpoints exposing full OpenAI Chat, Audio (/v1/audio/*), and Image Generation (/v1/images/*) API parity.
  • OpenTail.Stingray.Cli: Standalone command-line tool (stingray run, stingray image, stingray tts, stingray serve).

License

OpenTail.Stingray is licensed under the MIT License.

Product Compatible and additional computed target framework versions.
.NET net10.0 is compatible.  net10.0-android was computed.  net10.0-browser was computed.  net10.0-ios was computed.  net10.0-maccatalyst was computed.  net10.0-macos was computed.  net10.0-tvos was computed.  net10.0-windows was computed. 
Compatible target framework(s)
Included target framework(s) (in package)
Learn more about Target Frameworks and .NET Standard.

NuGet packages (1)

Showing the top 1 NuGet packages that depend on OpenTail.Stingray:

Package Downloads
OpenTail.Stingray.Server

ASP.NET Core endpoints, options, and DI extensions that expose OpenTail.Stingray as OpenAI- and Anthropic-compatible HTTP APIs. Bring your own host (Kestrel, IIS, YARP, etc.).

GitHub repositories

This package is not used by any popular GitHub repositories.

Version Downloads Last Updated
1.0.9 237 10/12/2026
1.0.8 135 10/10/2026
1.0.7 172 9/1/2026
1.0.6 151 8/19/2026
1.0.5 138 8/19/2026
1.0.4 141 8/12/2026
1.0.3 152 8/8/2026
1.0.2 155 8/6/2026