# Third-Party Notices — Neural Core

Neural Core itself © 2026 Tyler Duncan. Neural Core bundles the third-party materials below. Full license texts are
in `Shared/Licenses/` and are shipped inside the app at `Contents/Resources/Licenses/`, viewable in
**Settings › About › Acknowledgements & Licenses**.

**Neural Core is powered by NVIDIA Nemotron 3.5 Lightning 30B-A3B, used under OpenMDW-1.1. Not endorsed by NVIDIA.**
("Neural Core" is both the app's name and the user-facing name for the bundled model below; it is not an NVIDIA product name.)

---

## 1. Bundled AI model — NVIDIA Nemotron 3.5 Lightning 30B-A3B (fit16gib GGUF)

| | |
|---|---|
| File | `NVIDIA-Nemotron-3.5-Lightning-30B-A3B-fit16gib.gguf` (16,922,476,480 bytes) |
| SHA-256 | `187858b04dccae82a8c6fbf8bc5f0a62cfedb21d2f5aef3b4589456b09b6cd75` |
| Obtained from | https://huggingface.co/Alberto-Codes/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-fit16gib-GGUF (revision `3776a7d66a065b7d036a0fc8c058ac36cde0f4b2`) |
| Base model | https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 (revision `ce38b6ab8b252b4b8ee7165b4605e93191cafd73`) |
| Quantization | Alberto-Codes, llama.cpp b10362 + vramfit; imatrix from bartowski/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-GGUF (not redistributed by Neural Core) |
| License | **OpenMDW License Agreement, version 1.1 (OpenMDW-1.1)** — https://openmdw.ai/license/1-1/ |
| Modifications by Neural Core | None. The file is redistributed byte-for-byte (verified by SHA-256 at build time). |

**Required notice (retained per OpenMDW-1.1):**

> Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
> Licensed under the OpenMDW License Agreement, version 1.1. A copy of the agreement is included
> (`Nemotron-model-OpenMDW-1.1-LICENSE.txt`, and the quantizer repository's identical copy
> `Nemotron-fit16gib-GGUF-repo-LICENSE.txt`). Model cards of origin are included under
> `Licenses/model-cards/`.

NVIDIA and Nemotron are trademarks of NVIDIA Corporation. They are used only to identify the bundled
model. Neural Core is not affiliated with, sponsored by, or endorsed by NVIDIA.

### 1a. Upstream synthetic-data generator licenses (precautionary)

NVIDIA trained the model partly on synthetic data generated by other models. NVIDIA's dataset cards note that a model trained on that data "may be subject to redistribution and use requirements in the Qwen License Agreement … and the DeepSeek License Agreement". **Neural Core doesn't include any Qwen or DeepSeek weights.** The notices below are given as a precaution, pending legal review. Full details and hashes are in `Licenses/UPSTREAM-GENERATOR-LICENSES.md`.

**Built with Qwen.** Synthetic training data generated with Qwen2.5-72B, Qwen2.5-72B-Instruct and Qwen2.5-Math-72B, licensed under the Qwen LICENSE AGREEMENT (release date September 19, 2024; `Qwen-LICENSE-AGREEMENT-2024-09-19.txt`).

> Qwen is licensed under the Qwen LICENSE AGREEMENT, Copyright (c) Alibaba Cloud. All Rights Reserved.

**DeepSeek-V3.** Synthetic training data generated with DeepSeek-V3, licensed under the DEEPSEEK LICENSE AGREEMENT, Version 1.0 (`DeepSeek-V3-MODEL-LICENSE.txt`). To the extent the bundled model is a "Derivative of the Model" under that license, its use is subject to the use-based restrictions in paragraph 5 and Attachment A. These are reproduced in `DeepSeek-Use-Restrictions-NOTICE.txt` and incorporated in the Neural Core EULA.

The other generators NVIDIA used are licensed under Apache-2.0 (most Qwen2.5/Qwen3/QwQ models) or MIT (DeepSeek-V3-0324, V3.2, R1, R1-0528, R1-Distill, V4-Pro). Copies are in `Licenses/synthetic-data-generators/`. Those licenses place no conditions on models trained with their outputs.

Qwen is a trademark of Alibaba Cloud and DeepSeek is a trademark of its owner. Both are named only to identify data origins. Neural Core is not affiliated with or endorsed by either.

## 2. llama.cpp / ggml — embedded `llama.framework` (Vendor/llama.xcframework)

- Release **b11913** (commit `7aca0d2ea8d92b56911233b822db01f9fbb1ae50`), official prebuilt
  `llama-b11913-xcframework.zip`, SHA-256 `4daf8f536aeec4f6fac98ef823f31c8c5e79978a83b55a47199076cbfdfe7eb5`.
- License: **MIT** — Copyright (c) 2023-2026 The ggml authors. (`llama.cpp-ggml-MIT-LICENSE.txt`)

Third-party code compiled into that framework (verified by symbol inspection of the macOS slice):

| Component | License | Notice file |
|---|---|---|
| stb_image v2.30 (Sean Barrett), used by libmtmd | MIT **or** Public Domain (Unlicense) — Neural Core elects MIT | `stb_image-LICENSE.txt` |
| YaRN RoPE scaling code in ggml (Jeffrey Quesnelle & Bowen Peng) | MIT | `YaRN-NOTICE.txt` |

Not present in the shipped framework (checked): miniaudio, nlohmann/json, cpp-httplib (those are
used only by llama.cpp's CLI/server tools, which Neural Core does not ship).

## 3. Swift packages

None. Neural Core has no third-party Swift Package dependencies. (XcodeGen, MIT, is a developer build tool
only and is not distributed.)

## 4. Apple frameworks and SF Symbols

SwiftUI, AppKit, Foundation, Metal, Accelerate are part of macOS. SF Symbols are used in the UI under
Apple's SF Symbols license terms (not in the app icon).