Analysis updated 2026-05-18
Chat privately with an on-device AI model with no internet connection needed after setup.
Use the project as a starting skeleton for building your own private AI iOS app.
Add new AI models to the app by following the one-line code example.
Learn how Apple's MLX framework runs language models on iPhone GPUs.
| kiraa-ai/project-onyx | bleeeet/termipet | hoobnn/fanfan | |
|---|---|---|---|
| Stars | 34 | 33 | 33 |
| Language | Swift | Swift | Swift |
| Setup difficulty | moderate | easy | easy |
| Complexity | 3/5 | 3/5 | 3/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires a physical iPhone 15 or newer, the simulator cannot run it.
Onyx is an iPhone app that lets you chat with an AI language model entirely on your device, with nothing sent to any external server. There is no account to create, no API key to obtain, and no subscription. Once the model file is downloaded, the app works completely offline. The app ships with a pre-configured model called Llama 3.2 1B Instruct, a compressed version of Meta's Llama model that weighs about 860 MB. You download it from within the app's Models tab. The underlying inference is handled by Apple's MLX framework, which runs the model on the iPhone's built-in GPU. Because of this, the app requires a physical iPhone 15 or newer: the simulator does not have the necessary hardware, and older iPhones do not have enough memory. The interface has three tabs: Chat, Models, and Settings. The Chat tab is a standard conversation interface where the AI's responses appear token by token as they are generated. The Models tab lets you download, activate, or remove models. The Settings tab lets you set a system prompt (a background instruction that shapes the AI's behavior) and toggle some developer options. The project is designed as a starting point for developers who want to build their own private AI iOS app. The README describes it as a skeleton with deliberate gaps: conversations do not persist between app launches, there is no custom design, and there is no crash reporting. Code examples in the README show how to add new models (one line) and how to add conversation history that survives restarts. The app requires iOS 17 or later and is built with Swift 6. It is open source under the Apache 2.0 license.
An iPhone app for chatting privately with an AI language model that runs entirely on your device, no server or account required.
Mainly Swift. The stack also includes Swift, MLX, iOS.
Use freely for any purpose, including commercial use, as long as you keep the copyright and license notices.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.