enchanted

Enchanted is an open-source project that provides client applications for chatting with private self-hosted language models on iOS, macOS, and visionOS. It is compatible with Ollama and supports multiple models such as Llama 2, Mistral, Vicuna, Starling, and more. Its goal is to provide a ChatGPT-like application interface, but connected to your own private model, thereby enabling an unfiltered, secure, private, and multimodal experience within the iOS ecosystem.

Main capabilities include: conversation history, text-to-speech (read aloud), including conversation history in API calls, dark/light mode, conversation history stored locally on the device, Markdown support (can display tables, lists, and code blocks well), prompt support for image attachments, the ability to specify a system prompt for each conversation, edit message content or submit messages using a different model, delete a single conversation or all conversations, macOS Spotlight panel (Ctrl + ⌘ + K). All features work offline.

How to use: Enchanted requires Ollama v0.1.14 or later. There are two usage scenarios:

1. If the Ollama server has public access: download the Enchanted app from the App Store, specify the server endpoint in the app settings, and you can start using it.
2. If Ollama is running on a personal computer: start the Ollama server and download the model; install ngrok and forward the Ollama server to make it publicly accessible, with the command being ngrok http 11434 --host-header="localhost:11434"; copy the generated forwarding URL (similar to https://b377-82-132-216-51.ngrok-free.app), and your Ollama server API can be accessed through that temporary URL; specify that server endpoint in the app settings, and you can start using it.

Use cases: suitable for users who want to interact privately with self-hosted language models on iOS, macOS, or visionOS devices, with particular emphasis on privacy, no filtering, and multimodal experience.

Limitations: you must run the Ollama server yourself; the app itself does not provide models; Ollama v0.1.14 or later is required; if the server is local, public forwarding is required through tools such as ngrok. The project author mentions that Jaz is a new iteration of this project.