AccountyCat

Product Introduction

1. Definition: AccountyCat is an open-source, privacy-first focus companion application for macOS. It functions as a context-aware AI productivity tool that sits in the menu bar and intelligently monitors user activity to prevent procrastination without blocking legitimate work.
2. Core Value Proposition: AccountyCat exists to solve the fundamental flaw in traditional website blockers and distraction-blocking apps: their inability to distinguish between productive and unproductive use of the same application or website. Its primary value is providing intelligent, private, and non-intrusive focus assistance by reading on-screen context, not just URLs or app names.

Main Features

1. Context-Aware Drift Detection: AccountyCat works by periodically reading the active application name, window title, and—only when necessary—a temporary screenshot of your screen. This data is analyzed in real-time by an AI model (either locally or via API) to determine if your current activity aligns with your stated focus goal (e.g., "writing an essay"). It differentiates between a YouTube tutorial for your task and a cat video compilation.
2. Privacy-Centric AI Modes: The product offers two distinct, privacy-preserving AI operation modes. The 100% On-Device Mode uses llama.cpp to run Qwen multimodal models (4B, 9B, 27B) locally on your Mac; no data ever leaves the device. The Bring-Your-Own-Key (BYOK) Mode allows users to connect an OpenRouter API key, enabling access to more powerful cloud models (like DeepSeek V4 Flash, Kimi) while enforcing Zero Data Retention (ZDR) policies, ensuring providers cannot log or train on your data.
3. Configurable Personality & Nudges: Users can choose from three AI companion characters (Mochi, Misty, Onyx), each with a distinct tone and color palette. Interventions are short, contextual nudges (e.g., "hey — is that helping with the essay?") that escalate gently if ignored, but the system is designed to avoid interrupting legitimate work, which is treated as a bug.

Problems Solved

1. Pain Point: The failure of block-list-based productivity tools. Traditional app blockers and website blockers work by denying access to a list of sites, which is ineffective because the same site (e.g., Reddit, YouTube, X) can be used for both work and procrastination. This leads to users constantly disabling the blocker, rendering it useless.
2. Target Audience: The primary user personas are knowledge workers, developers, writers, students, and researchers on macOS who engage in deep work but struggle with context-switching and digital distraction. It specifically appeals to privacy-conscious users, developers who prefer open-source software, and Apple Silicon Mac users seeking efficient on-device AI applications.
3. Use Cases: Essential for scenarios like: writing a report or essay (distinguishing research from scrolling), coding (differentiating Stack Overflow lookup from browsing tech news), studying with online materials (keeping tutorial videos open while blocking unrelated content), and managing focused work blocks where intent matters more than the specific app being used.

Unique Advantages

1. Differentiation: Unlike blanket blockers (Freedom, Cold Turkey) or simplistic timer apps, AccountyCat uses multimodal AI to understand intent and content. Unlike surveillance-style employee monitoring software, it is user-centric, private, open-source, and designed for personal accountability, not reporting.
2. Key Innovation: The product's core innovation is its pragmatic use of on-device and ZDR-enforced cloud AI for real-time screen context analysis at the user level. By combining local LLM inference via llama.cpp with a BYOK cloud option, it delivers sophisticated reasoning without compromising privacy. The "screenshot-and-discard" architecture, where no visual data is permanently stored, is a critical technical and privacy design choice.

Frequently Asked Questions (FAQ)

1. Is AccountyCat really private? Yes, AccountyCat is architected for privacy by default. In on-device mode, all processing happens locally via llama.cpp. In cloud mode, using an OpenRouter key enforces Zero Data Retention. The app is open-source (MIT licensed), has no telemetry, and screenshots are analyzed and immediately discarded, never stored or uploaded to unknown servers.
2. How does AccountyCat differ from a regular website blocker? Standard website blockers use static block lists that cannot understand context. AccountyCat uses AI to analyze what is actually on your screen—the content of a YouTube video, the text in a Reddit thread—to determine if it's related to your work. It allows productive use of typically "blocked" sites and only intervenes when it detects a genuine drift in focus.
3. What macOS permissions does AccountyCat need and why? It requires two permissions: Screen Recording (to capture temporary screenshots for context analysis) and Accessibility (to read the active application and window title). These are standard for apps that interact with other app content. The open-source code allows anyone to audit exactly how these permissions are used.
4. Can I use AccountyCat without an internet connection? Absolutely. When configured to use the fully local, on-device AI mode (the default with downloaded models), AccountyCat requires no internet connection whatsoever after initial setup and model download, making it perfect for use on planes or in offline environments.
5. What is the cost of using AccountyCat? The application itself is free and open-source forever. Costs arise only if you choose the cloud AI mode via OpenRouter, where you pay for model inference (estimated $1–$5 monthly). The on-device mode using local Qwen models has no ongoing cost. A free trial code (ACFIRST) provides $0.50 of cloud credit to start.