VM API and SDK v3

SDK 3 uses VM, VMHandle, VMData, client.vms, client.vm(id) and oblien/vm. This is a breaking release and requires the matching VM API and gateway. Upgrade customers after that service release is deployed and verified. SDK 2 is incompatible with the new service; there is no legacy compatibility layer.

const vm = await client.vms.create({ image: 'node:22', wait_ready: false });
const { vms } = await client.vms.list();
await client.vm(vm.id).start();

REST clients use /vms. Other resources use native fields such as vmId, vm_id, source_vm_id and max_vms. Token scopes use vm; VM permissions use vm.*. Provisioning streams send event: vm and native VM envelopes. Platform machine codes use VM_*.

Old resource names, routes, response aliases, SDK exports and protocol aliases are removed. There is one native transport and no version selector. Each operation sends its native request once. CLI commands are oblien vm and oblien vms; use --vm-id, stop-vms and MCP vm_* tools.

The prepared runtime gateway is vm.oblien.com. API, gateway, scoped credentials, dashboard and stored VM identifiers require a coordinated release. Publishing SDK 3 alone does not activate routes. Customers must wait for the matching service release before upgrading.

Customer metadata, names, config, guest environment, labels and file/log contents remain opaque. Guest /workspace remains a filesystem directory. Verify signed webhooks against the original body bytes.

Customer upgrade checklist

1. Install oblien@3 and update SDK calls, REST paths, CLI commands and MCP tools to the native VM contract. Change platform-owned fields only; customer data remains opaque.
2. Update token issuance to the vm scope and vm.* permissions. Obtain fresh scoped and runtime credentials after cutover, and update event consumers to the native VM contract. Old signed tokens are not converted in place.
3. Coordinate the client switch with the matching API and gateway deployment. Stored VM IDs are preserved; existing VMs do not need to be recreated.
4. Verify an existing VM, a new disposable VM, runtime exec/files and account and namespace isolation before resuming customer traffic.