What is Copilot vs Agent?

A copilot is an assistant riding along: it lives inside an app you're already operating, watches your context, and offers the next line of code, the next sentence, the next formula. You accept or ignore it, and you do the work. An agent is handed the outcome instead — reconcile these invoices against the bank export — and goes off to do it across whatever tools that takes, returning with the result rather than a suggestion. The difference is delegation, not intelligence.

It follows that they fail differently, and need different safety. A bad copilot suggestion costs you a glance; a bad agent run costs you an email you can't unsend — which is why agents need approvals and copilots mostly don't. Velaris is built for the agent side: multi-step agents that reach across connected tools on their own, with destructive actions gated on human approval and credentials scoped per run. The desktop is where you brief and review them, not where you supervise each keystroke.