An AI chat app that actually wants you to read its source code

What it does FlowDown is a Swift/UIKit chat client for iPhone, iPad, and Mac that connects to OpenAI-compatible APIs or self-hosted inference servers like Ollama and LM Studio. It handles vision, audio, file attachments, web search, and prompt templates, with iCloud sync across devices. The app is moving to a free model this year, with subscriptions reserved for future personalization features.

The interesting bit The privacy posture is unusually aggressive: zero telemetry, zero crash reports, zero data retention by the developers. Your content stays on-device or in your private iCloud. The AGPL-3.0 license means the code is auditable, and the team has even open-sourced their chat UI component separately as LanguageModelChatUI. It’s the rare app that treats “you can verify what it does” as a feature worth advertising.

Key highlights

• Native Swift/UIKit build with Metal/MLX acceleration for on-device inference
• .fdmodel file format for importing and exporting model configurations in bulk
• Deep system integration: Shortcuts, Live Activity, default translation app slot
• Self-hosting first: Ollama, LM Studio, LocalAI, or any OpenAI-compatible endpoint
• Markdown rendering, syntax highlighting, and “buttery-smooth” interface (their words, but the UIKit-from-scratch approach suggests they earned it)

Caveats

• The app is explicitly “designed for users who want full control” and are “willing to configure them manually” — this is not plug-and-play
• The FlowDown name, icon, and artwork are proprietary despite the AGPL code; commercial use requires separate licensing
• Website uses Cloudflare and anonymous analytics, a minor asterisk on the zero-data-retention claim

Verdict Worth a look if you’re an Apple-ecosystem developer or power user who self-hosts models and resents the usual chat-app data hunger. Skip it if you want managed models, cross-platform support, or a setup that doesn’t involve manual endpoint configuration.