Skip to main content

OpenCLI

An agent for the models you run yourself​

Runs on your computer, works with any OpenAI-compatible model, and does the work rather than describing it: reads your files, runs your commands, edits your work, and keeps going in the background when you are not there.

Get it running​

Download OpenCLI-macos-aarch64.dmg (Apple Silicon) or OpenCLI-macos-x86_64.dmg (Intel), then drag it to Applications.

The build carries no Apple certificate, so the first launch needs one of these:

xattr -dr com.apple.quarantine /Applications/OpenCLI.app

or right-click the app and choose Open. Only once.

It ships with no model. Point it at one — a local Ollama, your own server, or any of forty-odd providers, several of which give a free allowance.

Where to go next​

First conversation

Where a chat starts, what it asks before changing anything, and how to see what it did.

Departments and bots

A directory with standing instructions, and chats inside it that have jobs. Nine ship with sample data.

Duties

Work that comes round on its own — and stops to ask you when it reaches something it should not decide alone.

Background runs

Work that keeps going after you close the chat that started it, and why one is sometimes held.

Sandbox and approvals

What the agent can do, and what it must ask about first. The most important page here.

What it cannot do

Said up front rather than discovered later, with the reason for each.

Two things worth knowing before you spend an evening on it​

A small local model is not as good at this as a frontier one. The interface is the same; the reasoning is not. And a model that will not call tools cannot do this work at all — it will chat. Check yours before committing to it.

The agent's sandbox can write anywhere inside its working directory. That directory is the whole of what it may change, which is why a conversation opens in a workspace rather than in your home folder.