Welcome to Quincy¶

Quincy is an AI assistant that runs on your Mac. It uses the same technology behind ChatGPT and Anthropic's Claude — large language models — but with a key difference: you stay in control.

Unlike cloud-only AI assistants, Quincy can run entirely on your machine. Your conversations, files, and personal data never leave your Mac unless you choose otherwise. And unlike assistants that have open-ended access to your computer, Quincy gives you fine-grained control over exactly what the AI can and can't do.

It's a Mac-assed Mac app — built with the same care and platform-native integration that Mac users expect from the tools they use every day. No Electron wrappers, no cross-platform compromises.

Why Quincy?¶

Your data stays on your machine¶

Quincy can run language models locally on your Mac using your own hardware — no internet connection required, no data sent anywhere. If a task needs more horsepower, you can connect a cloud provider like Anthropic's Claude for faster, more accurate responses. You choose what stays local and what goes to the cloud, on a per-task basis.

The AI only does what you allow¶

Most AI assistants have broad, open-ended access to your computer. Ask one to "clean up my inbox" and it might delete emails you wanted to keep, or reply to people on your behalf, or do something you didn't expect — because nothing stopped it.

Quincy works differently. Each task is handled by a specialist agent with a specific set of tools and permissions. An email agent might be allowed to read your inbox and suggest calendar appointments based on what it finds — but it literally cannot reply to anyone, delete messages, or take any action you haven't explicitly allowed. These aren't suggestions the AI promises to follow; they're hard limits enforced before any action is taken. The AI can't override them, even if it tries.

Predictable behavior in an unpredictable system¶

AI models are non-deterministic — ask the same question twice and you might get two different answers. This is fine when you're brainstorming, but it's dangerous when the AI is taking actions on your behalf. A slight variation in how the model interprets "archive old files" could mean the difference between tidying up your Downloads folder and deleting your tax returns.

Quincy addresses this by putting deterministic controls around the non-deterministic AI. The model decides what to do; the tools enforce how it's allowed to do it. Date ranges, folder restrictions, which actions are permitted — these are set in code, not in a prompt the AI might misinterpret. The AI's creativity stays in the planning; the guardrails stay in the execution.

Local models or cloud models — or both¶

You can run Quincy with models on your own hardware (via llama.cpp), with cloud models like Anthropic's Claude, or with a mix of both. The hybrid approach is powerful: use local models for routine, private tasks (free and fast), and cloud models when you need stronger reasoning — like planning a road trip or organizing a birthday party with fewer mistakes.

An MCP firewall for your tools¶

Beyond managing your own agents, Quincy acts as a firewall between external AI tools and your data. If you use tools like Claude Code, Cursor, or other MCP-compatible agents, you can connect them to Quincy's MCP servers — but Quincy controls exactly what they see. Bridge an external MCP server into Quincy, apply tool-level access rules, then re-expose only the tools you want via the scoped MCP registry. External agents see exactly the tools you've allowed — nothing more. See Extending Quincy with MCP for details on setting up bridges and controlling external access.

Built for your Mac¶

Quincy uses the macOS Keychain to store your passwords and API keys — the same secure storage that Safari and other Mac apps rely on. Configuration files are cryptographically signed so nothing can tamper with your agents' behavior without Quincy noticing. It's built on the same security infrastructure your Mac already provides.

What You'll Find Here¶

• Getting Started — Set up Quincy and run your first conversation
• Security & Trust — How Quincy protects your data and configuration
• Setting Up Anthropic — Connect to Anthropic's Claude for cloud-powered accuracy
• Setting Up llama.cpp — Run language models locally on your Mac
• Server & Clients — Use Quincy across multiple devices
• Agent System — How Quincy breaks tasks into focused, controlled pieces
• Choosing Models — Pick the right models for your needs
• Extending Quincy with MCP — Connect agents to external tools and data sources

Developer Documentation¶

Looking for the API reference? See the QuincyCore Developer Documentation.