Running it yourself

Self-hosting is the point of the project, so this page is about setting expectations accurately rather than selling you on it.

What it expects

• ·A container runtime — the services are designed to run as stacks, not as one process
• ·A GPU if you want local models, local embedding and video generation; otherwise expect to use hosted providers
• ·A vector store and a database, both included in the stacks
• ·More than one machine, realistically: we run the spaces on a Linux server and the core runtime and GPU work on a workstation

Your data stays yours

The knowledge vault is plain files on your machine. Generated code goes to your own git. The routing graph and the code index live with your installation. Model calls go wherever you configure them — including to a local runner, if you would rather nothing left the machine at all.

That is a statement about what you can configure, not about our own deployment. In our own operation most agent work goes to a hosted model provider; local runners are used for embeddings and a few agents. What a provider does with what you send it depends on that provider and your terms with them.

There is no telemetry in the core runtime. A few auxiliary components can send error reports or traces, but only if you set the corresponding configuration. This statement is about the software you run, not about this website.