Skip to content

Deployment options

Deployment options

@norbital-ai/pod is the authoring SDK and runtime. Deploying a workspace means getting its source built, its database provisioned, and its bundle served. Norbital Core is the managed platform that does all of this for you.

Core (managed)

Core operates the full workspace lifecycle for each organization:

  • Workspace Studio — primary authoring and release surface via the preset API (docs)
  • Sandboxed builds — Vite builds run in isolated Daytona sandboxes on Commit and merge-to-live; no build box to run locally
  • Checkpoints — immutable releases managed from Workspace Studio; preview and live environments, merge to live, rollback
  • Per-tenant databases — Neon Postgres provisioned per org, with schema migrations validated on a preview branch before touching live
  • Agent — in-workspace AI assistant (docs)
  • Source store, auth, billing, and multi-tenant operations

Builder / Executive / User licensing plus metered storage and AI. Enterprise adds a forward-deployed Norbital engineer.

Host-agnostic by design
Pod declares its dependencies as facility ports — Postgres, object storage, and an ai provider — rather than hard-wiring them. Core is the reference host that implements those ports. The same workspace source is portable because only the host bindings change, never the application code.

How a deploy works on Core

The primary path is Workspace Studio — builders edit source through the preset API, not by pushing to a separate org git remote. (Git storage exists on the host for versioning and checkpoints; authors interact through Studio.)

  1. You edit workspace source in Workspace Studio on a preview branch (applyWorkspaceFileOps).
  2. Commit runs preview sync: sandbox build, DDL validation on the branch's Neon DB, and a new preview checkpoint (branch_db_map).
  3. Merge to live rebases onto main, runs the unified deploy (build when needed, DDL validation, live migration), and advances the live checkpoint (tenant_workspace_head).
  4. If any step fails, the previous checkpoint keeps serving — deploys never break live operations.