Open-source software in Yah AI
Yah AI runs on the device because other people published the engines that make that possible. This page names each one and the license it is used under. The website itself uses no third-party code at all.
What is inside the app
| Component | What it does in Yah AI | License |
|---|---|---|
| llama.cpp | Runs every on-device chat model (GGUF) and the vision projectors. Built into the app as a binary framework from the llama.cpp source. | MIT — Copyright (c) 2023–2024 The ggml authors |
| sherpa-onnx | Runs the Kokoro neural voice on iPhone and iPad. Not included in the Mac build, which uses Apple's built-in voices instead. | Apache License 2.0 |
| ONNX Runtime | The inference engine sherpa-onnx runs on. Included with it, on iPhone and iPad only. | MIT — Copyright (c) Microsoft Corporation |
| Kokoro-82M | The neural voice model itself: a download of about 125 MB the first time you turn the voice on, never bundled in the app. | Apache License 2.0 |
| ml-stable-diffusion | Apple's Core ML runtime for Stable Diffusion: every on-device image, and the seed frames of on-device video. | MIT — Copyright (c) 2024 Apple Inc. |
| swift-argument-parser | A dependency of ml-stable-diffusion. | Apache License 2.0 |
Everything else the app is built on is Apple's own: Core ML, Vision, Speech, HealthKit, StoreKit, MultipeerConnectivity, FoundationModels and the rest of the iOS and macOS SDKs, under Apple's developer terms.
Downloaded models
The app ships with no AI model inside it. Every chat, vision and image model is a file you download from Hugging Face on first use, and each one is published by its own author under that author's license, which the app does not change. The license is on the Hugging Face page the file is fetched from, and the in-app catalog names it where the publisher states it plainly — Microsoft's Phi-4 Mini Reasoning, Phi-3.5 Mini and MediPhi-Instruct, for example, are MIT-licensed. The Stable Diffusion checkpoints are community models converted to Core ML; their terms are their publishers'. Read the page for a model before you download it if what you make with it matters commercially.
The MIT License
Applies to llama.cpp, ONNX Runtime and ml-stable-diffusion, with the copyright holders named above.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
The Apache License, version 2.0
Applies to sherpa-onnx, the Kokoro-82M model and swift-argument-parser. Each is licensed under the Apache License, Version 2.0 (the "License"); you may not use it except in compliance with the License. Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. The full text is at apache.org/licenses/LICENSE-2.0.
Questions about a license or a missing attribution: support.