Agenteam

Agenteam is a macOS app that takes the coding agent command line tools you already have installed and gives each one a body on your desktop. Claude Code, Codex, opencode and pi become little characters that float above every other window. Click one and a chat balloon opens beside it, on whichever side of the screen has room, and you tell it what you want in the same way you would in a terminal. It does the work inside the folder you gave it, then reports back in the balloon.

The problem it's solving is small but real. Agent CLIs are good, and they're also invisible. They live in a terminal tab you have to go find, they lose their thread when you close the window, and running three of them for three different jobs means three tabs you keep confusing. Agenteam's answer is to stop treating an agent as a session and start treating it as a thing that exists. There's no window to hunt for and no tab to switch to, because the agent is already on top of whatever you're doing. The character sitting on your desktop is also a reminder that something is running, which is a mild but genuine benefit over a process buried three windows deep.

Each agent gets one workspace, either a folder of its own or a repo you point it at, and the file tree sits next to the conversation so whatever it just wrote is one click away. The model here is one agent per thing you care about, so you might keep one in your notes, one in a repo and one watching something else, each with its own scope and its own history. Agents also keep their memory, since each holds a single long conversation that survives restarts, and the desktop balloon and the main window are two views onto the same thread rather than separate chats. That continuity is the practical difference from a terminal session, where closing the window usually means starting the explanation over.

Agents can schedule themselves. Ask for something every morning at nine and the agent writes itself a routine, runs on that schedule, tells you what happened afterwards, and can be switched off in the same place you set it up. That's the closest the app gets to autonomy, and it stays deliberately visible rather than running quietly in the background where you'd forget about it. Permissions are set per agent, with a choice between asking before it acts, planning only, or never asking, and the app warns you if you point an agent at your entire home folder instead of a scoped workspace.

What makes it different from most agent products is that it doesn't sell you inference. Agenteam runs no model and resells no tokens. It drives the CLI you've already installed and paid for as a long-lived process on your own Mac, so whatever signs you in at the terminal, a subscription or an API key, is what the app uses, and it adds no account of its own. Conversations, settings and transcripts sit in the app's own folder on your machine. Whatever the CLI you picked sends upstream is between you and that CLI, which is a narrower privacy claim than most apps make and a more accurate one. The licensing works the same way, since the key is verified with the developer once and checked locally after that, so the app keeps working offline and doesn't call home again.

It suits developers who already live in a coding agent CLI and want several of them running against different projects without drowning in terminal tabs. The requirements are honest about the boundaries, since you need one of the four supported CLIs working in your terminal first, and the app is Apple silicon and Intel Macs on a recent macOS with no Windows or Linux build yet. It lives in the menu bar rather than the Dock, so it stays out of the way when you aren't using it. The app itself is closed source, described by its author as a closed-source app that drives open tools, and what you download is the built, signed and notarised binary. Support is a direct email address on the site, which early on reaches the person who wrote the thing rather than a ticket queue.

Pricing is unusually simple. The free version is the whole app with a cap of three agents, all four backends, routines, workspaces, themes and every update, with no account, no clock and no trial that expires. The reasoning given is that three is enough to find out whether you want a fourth. Lifting the cap costs a one-time early-bird payment of $9.9, which covers as many agents as you like on up to three Macs at a time, with slots you can release from Settings when you retire a machine. The licence is checked once and then verified locally from then on, so it works offline and doesn't phone home. The site says that price is an early-access number and will rise once things settle.