Skip to main content
Popular

Build Your Own AI Desktop Agent

Learn to build a real AI agent that runs on your own hardware, connects to Slack, sees your screen, and keeps working when you close your laptop.

Updated Sep 14, 2026

About this course

Most AI agent tutorials run in a cloud sandbox that disappears when the session ends. This course is about something different: a persistent agent that lives on a Mac Mini, survives reboots, responds to Slack mentions, and can see and interact with your actual desktop. That's a different class of tool, and it requires a different approach to build. You'll work through the full stack from the ground up. Hardware setup and macOS configuration. A Node.js agent core using the Vercel AI SDK with real tool-calling. An orchestrator that routes events and spawns short-lived workers for parallel tasks. A Cloudflare Tunnel that makes your Mac reachable over HTTPS without opening firewall ports. A Slack app built from a JSON manifest with proper HMAC signature verification. And finally, screenshot and click tools that give your agent eyes and hands on the desktop, running as a persistent launchd service. Every lesson builds toward something you can test immediately. By the end, you'll @mention your bot in Slack, it will screenshot your desktop, and reply in-thread with the image. No managed products, no wrappers, no skipped steps. You'll know exactly how every piece works because you wrote it.

Details

Updated Sep 14, 2026
7 units, 14 lessons
7 projects
7 assessments

Skills you'll learn

Persistent Agent Architecture

Design and build a five-layer desktop agent stack that runs continuously on local hardware instead of disappearing after a session ends.

Tool-Calling with the Vercel AI SDK

Define typed tools using Zod schemas and wire them into a streaming agent loop that calls bash, reads files, and writes output.

Orchestrator/Worker Patterns

Split agent work between a long-lived orchestrator that routes events and short-lived workers that execute scoped tasks in parallel.

Slack App Integration

Build a Slack app from a JSON manifest, verify webhooks with HMAC-SHA256, and route inbound mentions into an agent inbox with threaded replies.

macOS Automation and Persistence

Grant the right system permissions, add screenshot and click tools to your agent, and install it as a launchd service that restarts on crash.

Syllabus

7 units • 14 lessons • 7 projects • 7 assessments

Learning activities

Every lesson gives you several ways to learn.

Read
Flashcards
Quiz
Comic

Frequently asked questions

Course thumbnail