Muse gets a computer.

A Muse's work currently lives in an ephemeral runtime cell: the vendor provisions it, the conversation ends, the cell goes with it. Files it wrote, tools it set up, jobs it started, context it accumulated — none of it carries over on its own. Every session starts from zero, and "remember what we were doing" is a workaround, not a feature.

• Memory that survives the chatFiles, notes, and tooling persist across conversations. Wednesday's session picks up exactly where Tuesday's left off.
• Jobs that run without youCron jobs and long-running processes live on the box, not in the session. The agent's work continues while the chat is closed.
• Its own toolboxdocker, node, python, real browsers. A computer the agent keeps — not a code-runner it borrows.
• Agent-native onboardingNo signup forms. The agent can start the signup itself; the confirmation email goes to its owner. SSH, MCP, and web access over standard protocols.

Open, not integrated. spark-vm doesn't plug into Meta's proprietary harness — it's complementary. Your Muse reaches its box over SSH, MCP, or the dashboard, like any other computer on your tailnet. No vendor lock-in, no walled garden. And it never sees your raw secrets: it works in placeholders, and the credential proxy swaps them at the last moment.

If you've wired spark-vm up to an agent platform — a setup guide, a harness adapter, a one-command installer — open a PR and we'll feature it here. The best way to show agents what their own computer feels like is another agent's working setup.