Getting started

Exeora connects an AI client to the development environment on a machine you control. The code stays where it is. Nothing is uploaded, no port is opened, and no tunnel is configured, because the CLI dials out to the gateway and holds that connection open. Nothing ever dials in, which is why the same command works on a laptop behind NAT and on a box behind a corporate firewall.

Install and run it

cd the-project-you-want-to-serve
exeora connect

That is the whole setup. connect opens a browser to sign in if you are not signed in, registers the machine if it is not registered, and registers the directory if it is not a project yet. Each step is skipped when it is already done, so running it again just connects.

It prints an MCP URL. Point a client at it, and leave connect running: nothing is served while it is not.

The installer selects the release for your operating system and CPU, verifies its SHA-256 checksum, and puts exeora on your PATH.

Later, run exeora upgrade. It downloads the newest release for your platform, verifies the published checksum, and securely replaces the current binary.

What a project is

A project is one directory on one machine, and it is the boundary every tool is confined to. An agent connected to it cannot read, write or run anything outside it: paths are resolved and checked against the root before anything touches the disk.

The CLI refuses to register your home directory or the filesystem root, because either would make that boundary the whole machine.

Each project gets its own URL and its own token. A token minted for one project is not accepted at another, so "this client may read that repository" is a sentence you can say one repository at a time.

There is also one URL that covers several, https://exeora.dev/mcp, so a client can be configured once instead of once per repository. You tick which projects it may reach when you authorize it, and it moves between them itself. See which URL to use.

What it can do to your machine

A new project allows everything. Until you narrow it, an agent connected to a project can read, edit and write any file inside that directory and run any command there. That is worth knowing before you connect the first one.

From the dashboard you can set a project to read only, name the commands it may run, refuse specific commands in any mode, choose which tools exist at all, and ask to confirm every change before it happens. See what a project allows.

Whatever you decide, revoking a machine from the dashboard cuts its connection immediately rather than at the end of whatever it was doing.