Build on alexandr

alexandr is a model-agnostic AI workspace built on a generic OS runtime. The same runtime runs on your Mac, in our cloud, or on your own server — and almost everything inside a workspace is an app or a skill you can build yourself.

Private preview. The source isn't public yet and the @alexandr/* packages aren't on npm. You can still build today against a local runtime — see the Quickstart.

What you can build

• Apps — an npm package the runtime builds and serves at /apps/<id>/. A manifest, a React entry, and the SDK is the whole contract. Start with App anatomy.
• Skills — a single SKILL.md the assistant installs into its brain from the catalog feed. See Skill anatomy.
• Kernel packages — go deeper and extend the runtime itself via boot(), adding routes, tools, or whole subsystems.

How it fits together

The runtime is one container holding one workspace, with its data in a volume you own. Apps render into the workspace and talk to the kernel through @alexandr/sdk; skills extend the assistant. The open catalog feed is where installable apps and skills come from — the runtime pulls from it the same way in every placement.

Next steps

• Quickstart — get a runtime running locally.
• Core concepts — the runtime, workspaces, apps, skills, and the assistant.
• App anatomy — build your first app.