---
license: mit
base_model: microsoft/MagenticBrain
tags:
  - onnx
  - onnxruntime-genai
  - qwen3
  - agentic
  - tool-calling
  - text-generation
---

# MagenticBrain ONNX (INT4)

This repository contains an ONNX INT4 conversion of
[microsoft/MagenticBrain](https://huggingface.co/microsoft/MagenticBrain)
for use with [ONNX Runtime GenAI](https://github.com/microsoft/onnxruntime-genai)
and the [ElBruno.LocalLLMs](https://github.com/elbruno/ElBruno.LocalLLMs) library.

## Model Description

MagenticBrain is a 14B-parameter orchestration model fine-tuned from Qwen3-14B by
Microsoft Research AI Frontiers. It is designed for planning, tool selection, multi-turn
tool chaining, and sub-agent delegation in agentic applications.

## Conversion Details

| Field | Value |
|---|---|
| Source | `microsoft/MagenticBrain` |
| Precision | INT4 (kld-block-128 quantization) |
| Execution provider | CPU (universal) |
| Tool | `onnxruntime_genai.models.builder` |
| Architecture | Qwen3-14B decoder-only |

## Usage with ElBruno.LocalLLMs

```csharp
using var client = await LocalChatClient.CreateAsync(new LocalLLMsOptions
{
    Model = KnownModels.MagenticBrain,
    EnsureModelDownloaded = true   // downloads automatically on first run
});
```

## License

MIT — inherited from the source model.