Quack
A native macOS AI chat client that connects to multiple LLM providers from a single interface, with built-in tools, MCP (Model Context Protocol) support, and on-device inference.
Features
- Multi-provider support -- Chat with models from OpenAI, Anthropic, Google Gemini, Vertex AI, Ollama, Apple Intelligence (on-device), MLX, OpenRouter, Groq, Together, Mistral, and any OpenAI-compatible endpoint.
- Built-in tools -- Read files, write files, run shell commands, fetch URLs, and search the web without any external setup.
- MCP integration -- Connect external MCP servers via stdio transport with a three-tier permission model (Always Allow, Ask, Deny) and per-session server selection.
- MLX on-device inference -- Download, load, and run MLX models from HuggingFace locally on Apple Silicon with no API key required.
- Assistants -- Create reusable presets that bundle a provider, model, system prompt, parameters, tool permissions, and MCP servers together.
- Chat management -- Persistent conversation history with session pinning, archiving, search, date-grouped sidebar, and per-session model/parameter overrides via the inspector panel.
- Streaming responses -- Live token streaming with reasoning/thinking model support, configurable reasoning effort, and collapsible reasoning display.
- Token usage and cost tracking -- Per-message and per-session token statistics with estimated cost calculation via models.dev pricing data.
- Transcript export -- Export any conversation as a Markdown file (Cmd+Shift+E).
- Markdown rendering -- Full CommonMark rendering of LLM output including code blocks, tables, lists, and more.
- Auto-generated titles -- Chat sessions are automatically titled using on-device Apple Intelligence.
- Secure credentials -- API keys stored in the macOS Keychain.
- Auto-updates -- Built-in update mechanism via Sparkle.
Requirements
- macOS 26.0 or later
- Xcode with Swift 6.0 support