Guides

Last updated: 2026-06-27

These guides are for doing the work, not memorizing flags. Each one starts with a task, shows runnable commands, and points to the CLI reference when the details matter.

Run things

• Run one-shot commands - boot a microVM, run a command, tear it down. Setup steps, env vars, artifacts, timeouts.
• Keep a persistent workspace - create a named workspace and walk the create, start, halt, connect, delete lifecycle.
• Run a service - run Postgres in a workspace with a published port, a named volume for data, and a restart policy.

Move data

• Use volumes and move data - named volumes, attached disks, tar bundles, and cp in and out of stopped workspaces.
• Deliver secrets - get credentials into the guest without writing them to disk, plus on-demand fetch and the audit log.

Save and share state

• Snapshot and fork workspaces - checkpoint a running workspace, restore it in place, or fork copies from one snapshot.

Connect things

• Networking - give a workspace outbound access and publish a guest port back to the host.
• Allowlist and passthrough egress - confine a workspace to a known set of destinations with strict, and let cert-pinned endpoints through with passthrough.
• Serve microagent over MCP - register microagent serve mcp in Claude Code or another MCP client and drive workspaces with tools.

Build an agent

• Build a simple agent - a one-shot agent that takes a request, calls Claude under operator-supplied constraints, and writes a result.
• Build agents on the mediation channel - the guest-to-host vsock contract: declare it, listen on the host, loop in the agent.