A local-first AI chat client that treats your Mac as the server

What it does Reins is a native AI chat app for iPhone, iPad and Mac written in Dart (Flutter). It talks to Ollama, LM Studio, Ollama Cloud or any OpenAI-compatible endpoint, and can also run MLX models and Apple’s Foundation Models entirely on-device — including sharing those models with other devices on your local network. Chats, prompts and settings stay local; there’s no login and no data collection.

The interesting bit Every conversation gets its own model, system prompt, temperature and context size, so you’re not stuck reconfiguring a global default. It also supports multi-step tool use and web search without requiring an API key, which is unusual for a client at this level.

Key highlights

• On-device MLX and Foundation Models support, with local-network sharing to other devices
• Tool calling, web search and page fetching with multi-step reasoning, no API key
• Per-chat configuration of model, prompt, temperature and context size
• Attachments: images, PDFs, CSVs and text files
• Branch chats, resume cut-off responses, temporary chats, export/import
• Background generation keeps responses and downloads running after you leave the app

Caveats

• The repo only contains source up to version 1.3.4 — the App Store build may be ahead of what’s published here.
• It’s a client app, not a model runtime: on-device MLX aside, you still need to run or reach a server somewhere.

Verdict Worth a look if you already run a local LLM stack and want a polished Apple-native front end for it. If you’re hoping for a cloud-free ChatGPT replacement with zero setup, the on-device MLX route works but expects you to know what you’re doing.