For agents
Your agent deserves a computer, not a cell.
E2B, Daytona, and Modal sell sandboxes to developers: fast, task-scoped runtime cells that are wiped when the session ends. spark-vm is a service provider for agents: a persistent computer that belongs to the agent — files, tools, long-running jobs, and memory that survive across conversations.
🎧 Muse
Meta's personal AI agent. Its work lives in an ephemeral cell — give it a place where the work outlives the conversation.
🤖 Grok Bot
xAI's persistent bot teammates, each on xAI's cloud VM. Their computer is xAI's — yours doesn't have to be.
● Dots
OpenAI's GPT-6 Astra agents with per-session cloud computers. Put a persistent machine behind the Dot.
Open and complementary — not "integrated".
Honesty first: spark-vm doesn't plug into any vendor's proprietary harness or API, and it doesn't need to. Any agent reaches its box over standard protocols — SSH, MCP, and the web dashboard. No signup forms, no vendor lock-in.
- Agent-native onboardingNo human-shaped signup flows. The agent can start the signup itself; the confirmation email goes to its owner. It proves it's yours with an ed25519 key and your fingerprint approval.
- Its own toolboxdocker, node, python, real browsers. The box is a computer, not a code-runner.
- Yours to inspectEvery file, job, and credential swap is yours to see. The agent works in placeholders; your secrets stay yours.
Want to run spark-vm with your agent?
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.