AICompleteChat is a full-source, production-quality example of a fully on-device AI chat app for macOS. No cloud dependency once the model is downloaded — chat, voice dictation, and memory all run locally, assembled from AIChatKit, AIChatKitMLX, AiPersona, AiVoiceKit, and a real design system.
Runs mlx-community/gemma-4-e4b-it-4bit, downloaded from Hugging Face on first launch and cached locally after that — every subsequent launch is fully offline.
AiPersona gives the app a bi-temporal knowledge graph that extracts facts and entities from conversations, retrieves relevant ones per turn, and lets you browse, correct, or delete anything it's learned from Settings → Memory. Set your name once and it becomes a real fact in the graph, not just a settings field.
AiVoiceKit provides on-device speech recognition, global hotkeys, and AI-assisted rewriting of selected text in any app.
Conversations persist across launches, with delete support, via a SwiftData-backed ChatHistoryStore.
The chat interface lives in DesignFoundationPro's AIChat vertical (DFAIChatRootView and friends) — the app wires real data into it rather than reimplementing chat UI from scratch.
git clone https://github.com/NerdSnipe-Inc/AICompleteChat.git
cd AICompleteChat
xcodegen generate
open AICompleteChat.xcodeprojGrab the signed, notarized v1.0.0 release build — no build requiredRequires macOS 15+ on Apple Silicon, Xcode 26+, xcodegen (brew install xcodegen), and a DesignFoundationPro license — the one intentional gate, since the UI is assembled from DesignFoundationPro's AIChat vertical rather than built from scratch. Every other dependency resolves publicly via Swift Package Manager.
The source is meant to be read and learned from by anyone — it demonstrates how AIChatKit, AIChatKitMLX, AiPersona, and AiVoiceKit fit together in a real app. It only actually builds for someone with a DesignFoundationPro license, since the UI is assembled from that package's AIChat vertical.
Note: the code here is MIT-licensed; DesignFoundationPro, AiPersona, and AiVoiceKit are each licensed separately under their own repos' terms.
No — the build requires it, since the entire UI is assembled from DesignFoundationPro's AIChat vertical. You can still read the source to see how the AIChatKit/AiPersona/AiVoiceKit stack fits together, or grab the signed v1.0.0 release build to try the app without building it yourself.
No — it requires Apple Silicon, since MLX doesn't run meaningfully on Intel.
No, after the initial model download. Inference, memory, and voice processing all run on-device.
Read the source to see how AIChatKit, AIChatKitMLX, AiPersona, and AiVoiceKit compose into a real app, or grab the signed release build to try it directly.