quincy.wtf
An AI assistant that runs on your Mac — privately, securely, and under your control.
Made for Mac users by a Mac user.
# Your data stays on your machine
Quincy can run AI models directly on your Mac — no internet connection required, no data sent anywhere. If a task needs more horsepower, you can connect a cloud provider for faster, more accurate responses. You choose what stays local and what goes to the cloud.
# The AI only does what you allow
Most AI assistants have open-ended access to your computer. Ask one to “clean up my inbox” and it might delete emails you wanted to keep.
Quincy works differently. Each task is handled by a specialist agent with a specific set of tools and permissions. An email agent might read your inbox and suggest calendar appointments — but it literally cannot reply, delete, or take any action you haven’t allowed. These aren’t suggestions; they’re hard limits enforced before any action is taken.
# Use it everywhere
Start a conversation on your Mac, pick it up on your iPhone. The server runs on your Mac (or a dedicated machine); the command line, macOS app, and iOS app are lightweight clients that connect to it. Your agents, models, and conversation history stay in one place.
# Extend with any tool
Quincy comes with built-in tools, but you can connect it to almost anything — task managers, calendars, email, smart home devices, databases. Ask Quincy to set up a connection and it walks you through it. You control which tools each agent can see and what actions require your approval.
# Under the Hood
For developers and tinkerers
llama.cpp + GGUF
Local inference via llama-server in router mode. Each agent can target a different model.
Anthropic Claude
First-class cloud provider. Use locally, in the cloud, or both — per agent.
MCP Protocol
Connect any MCP server. Bridge external tools with approval policies and scoped access.
HMAC-Signed Configs
Every config is cryptographically signed. Keychain-backed root key with HKDF derivation.
macOS Keychain
All secrets stored in the Keychain. The LLM never sees your credentials.
Native Swift
Built with SwiftUI and Hummingbird. No Electron. No cross-platform compromises.
# Documentation
Getting Started
Onboarding, first-run setup, and configuring your first agent.
llama.cpp Setup
Install and configure local LLM inference via llama-server.
Anthropic Setup
Connect Quincy to Claude for cloud-powered accuracy.
Agent System
How the orchestrator and sub-agents work together.
Choosing Models
Model sizes, quantization, and the hybrid approach.
Security & Trust
Config integrity, Keychain secrets, and LLM key isolation.
Looking for the API? See the QuincyCore API Reference.