Skip to main content
Skip to main content

Remote or local setup

Use this page to decide where the agent workspace lives: inside Zerops, or on your machine beside the local app directory.

Both setups connect the agent to one Zerops project and the same project-scoped operations. The tradeoff is where the agent works: filesystem ownership, private-service access, credential location, bundled tooling, and shell blast radius. The runtime layout is a separate app-work decision.

What you are choosing

Remote setup always includes the zcp@1 workspace service. If the chosen runtime layout has a development runtime, the agent works with that service too.

Local setup does not need the workspace service or a separate development runtime. It runs beside your local source tree and deploys to the Zerops runtime you choose to link.

Remote setup is the safer default for broader agent autonomy because the agent shell, private networking, and workspace files stay inside Zerops instead of on your machine. Use local setup when local files, data, desktop tools, git credentials, or a local-only agent client need to stay local.

Setup comparison

Remote setupLocal setup
Workspace serviceAlways has zcp@1.Not required.
Development runtimeOnly if the runtime layout includes one.Not required for local work.
zcp processRuns in Zerops.Runs on your machine.
Agent processRuns in the remote workspace.Runs in your local editor or CLI.
Files the agent editsWorkspace files and mounted runtime files.Files on your machine.
Private service accessPrivate Zerops network from the workspace.Zerops VPN from your machine.
Token storageInjected into the workspace service..mcp.json in the local app directory.
Git credentialsConfigured inside the workspace service.Your local git credentials.
Safety postureBroad agent shell permissions stay in Zerops.Agent shell permissions affect your machine.

Starting points

Remote setup can start from:

  • Recipe with AI Agent environment. A guided path for a known stack. A recipe creates app services, managed services, and the zcp@1 workspace together.
  • New Zerops setup with remote setup enabled. Start from blank services or a custom stack and add the zcp@1 service during creation.
  • Existing development or staging setup. Add the zcp workspace next to services that already exist. Do not add ZCP to production; promote verified work through your release process instead.

Local setup can start from:

  • Empty local directory. The agent creates the app structure from a product request and uses ZCP to select or create Zerops services.
  • Existing app directory. The app code, editor setup, local data, and git credentials already live on your machine.
  • Recipe prepared for local setup. The recipe creates the Zerops service baseline, while the agent and files stay local.

Runtime layout is separate

Remote or local only answers where the agent and zcp process run. The project can still use:

  • one mutable dev runtime,
  • a dev + stage pair,
  • a single app runtime,
  • local files linked to a stage target.

That choice belongs to app work: Build and ship.

Stage is not production. Keep production in a separate Zerops project and promote work through your release process; see Production boundary.

Switching later

Remote and local setup are not permanent, but switching is a handoff between workspaces, not a sync operation. Before switching, make the handoff explicit:

  • Preserve the current source tree through git or another explicit copy path.
  • Choose one deploy source for the next task: the remote workspace, mounted runtime files, or the local app directory.
  • Do not edit mounted runtime files and a local repo in parallel unless you have a merge plan.
  • Regenerate local .env snapshots after Zerops env changes.
  • If the local deploy target changed, have the agent inspect the current local link before deploying.

Next steps

  • What remote workspace gives you - remote setup, Browser VS Code, bundled agent CLI, and workspace persistence.
  • Run locally - local install, zcp init, .mcp.json, VPN, env snapshots, and deploy target linking.
  • Trust model - how the safety boundary changes between remote setup and local setup.