← Back to Projects
🤖

Atrak Agent

OS-level AI assistant for macOS and Windows that integrates multiple LLMs and APIs to automate tasks, manage workflows, and boost productivity—all running locally on your machine.

Status: Active Development Started: Sep 2025 Lead: Charlie Han Platforms: macOS, Windows

🎯 The Problem

Modern knowledge workers spend hours on repetitive tasks: scheduling, file management, research, and context-switching between apps. AI assistants exist, but they're cloud-only and lack deep OS integration.

❌
Cloud Dependency

Most AI tools require constant internet and send all your data to external servers.

❌
Shallow Integration

Chat interfaces can't control your desktop, open apps, or manage files directly.

❌
Context Loss

Every new session forgets your workflows, preferences, and ongoing projects.

👥 Users & Impact

Local
Privacy-first, runs on device
Multi-LLM
Routes to best model per task
Plugins
Extensible tool ecosystem

Target Users

  • Developers wanting AI-assisted coding and automation
  • Researchers managing large knowledge bases
  • Power users automating repetitive workflows
  • Teams needing local-first AI for sensitive work

🎬 Demo

Agent Preview

🌙

Command bar + task queue workflow

Preview of the local-first desktop automation flow.
Atrak Agent workflow preview

Interface Preview

🚀 How to Run

🧪Alpha Access
💻From Source

Join Alpha Testing

  1. Fill out the access request form
  2. We'll review and send you a download link
  3. Install and configure your API keys
  4. Join our feedback channel for updates

Alpha builds are available for macOS and Windows. Linux is planned.

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                       Atrak Agent                           │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐  │
│  │   Command    │───▶│   Router     │───▶│   LLM Pool   │  │
│  │   Parser     │    │   (intent)   │    │   (GPT/Claude│  │
│  └──────────────┘    └──────────────┘    │   /Local)    │  │
│                                          └──────────────┘  │
│         │                                       │          │
│         ▼                                       ▼          │
│  ┌──────────────┐    ┌──────────────┐    ┌──────────────┐  │
│  │   Plugin     │◀──▶│   Safety     │◀──▶│   Response   │  │
│  │   System     │    │   Layer      │    │   Handler    │  │
│  └──────────────┘    └──────────────┘    └──────────────┘  │
│                                                             │
├─────────────────────────────────────────────────────────────┤
│  OS Layer: File I/O │ Process Control │ Clipboard │ Shell │
└─────────────────────────────────────────────────────────────┘
                            

Tech Stack

Python Electron GPT-4 Claude Local LLMs

✨ Key Features

⌨️

Quick Command Bar

Global hotkey summons an AI-powered command interface from anywhere.

🔀

Smart LLM Routing

Automatically picks the best model for each task—fast for simple, powerful for complex.

🔌

Plugin System

Extend with custom tools: calendar, email, code runners, web scrapers, and more.

🛡️

Safety Constraints

Configurable guardrails prevent destructive actions without confirmation.

💾

Context Memory

Remembers your projects, preferences, and ongoing tasks across sessions.

🔒

Local-First

Your data stays on your machine. API calls are minimal and can use local models.

🖥️

Cross-Platform

Designed to run on both macOS and Windows with the same workflow and UI.

📊 Metrics & Results

~200ms
Command parse time
12
Built-in plugins
3
LLM providers supported
macOS + Windows
Supported platforms

Development Notes

Currently in alpha testing with 5 internal users on macOS and Windows. Focus is on stability and core workflow coverage before expanding to Linux.

🗺️ Roadmap

✓
v0.1 — Core Agent

Command parsing, LLM integration, basic file operations

●
v0.5 — Plugin System

Extensible tools, calendar/email integration

○
v1.0 — Public Alpha

Stable release, documentation, onboarding flow

○
v2.0 — Multi-Platform

Linux support, local LLM optimization

🙏 Credits

Project Lead

Charlie Han

Development

Atrak Team

AI/ML

OpenAI, Anthropic, Ollama

Framework

Python + Electron

Inspired by the vision of truly helpful AI assistants that respect user privacy and agency.