Naseem for macOS - A native AI agent that does real work on your Mac

Naseem is a native AI agent for macOS, built entirely in Swift. It can work with your files, run terminal commands and Python, drive native Mac apps, automate and test apps in the iOS Simulator, use MCP and Skills, delegate to sub-agents, and work remotely through Telegram. Bring your own model or run locally with Ollama/MLX. Every consequential action requires your approval first, and autonomous mode is strictly opt-in. No Electron. No Node. It does real work on your Mac — with you in control.

Hi Product Hunt 👋 I built Naseem because I wanted an AI agent that doesn’t just live in a chat window — I wanted one that actually lives on my Mac and can do work there. So I built it natively in Swift. No Electron, embedded browser, or Node runtime. Naseem can work with files, use the terminal and Python, control native Mac apps, drive the iOS Simulator, use MCP servers and reusable Skills, delegate to sub-agents, and even be reached remotely through Telegram. One of my favorite tests was giving it an empty Xcode project and asking it to build a small Minecraft-style iOS game. It wrote the code, hit build errors, fixed them, rebuilt, launched the Simulator, interacted with the game, and tested what it had made. That’s the direction I want to take Naseem: LLM + Agent Harness + Your Mac You choose the model — cloud APIs, Ollama, or local MLX — while Naseem provides the tools and environment that let the model actually act. Naseem Free is free forever, and every install includes a 30-day Pro trial with no card required. This is still the beginning, and feedback is one of the main reasons I’m launching it here. I’d especially love to know: what real task would you hand to an agent that lives on your Mac? — Ayman Hamed

Replies

Best

Naseem for macOS

Hunter

📌

2mo ago

Dial

the Xcode/Minecraft-clone test is a great demo of the loop actually closing (build, fail, fix, rebuild). since it's driving the terminal and native apps directly on your own Mac, is there any confirmation step before something destructive - deleting files, force-pushing, running a script that touches more than the test project - or does it just act and you review after the fact?