Multi-Hub Topology

Hubitat's two planes behave differently across a multi-hub setup. Conflating them is why "I deployed to the hub" sometimes hits the wrong one.

Code is per-hub; devices can mesh

• Code (apps, drivers, libraries) lives in each hub's own editor. There is no mesh for code — deploy, pull, and log-tail target one hub by IP. A driver installed on hub A does not exist on hub B until deployed there too.
• Devices can mesh: with hub meshing, one Maker API instance can expose devices from secondary hubs. Device control can be centralized; code deployment cannot.
• Hub connection details are keyed by hub, one IP each; the hub-config skill owns that config.

Local network, no Hub Security (assumed default)

• These rules and the deploy/log-tail mechanisms assume hubs on the local network with Hub Security off — no login, no cookie, direct IP access (verified baseline in skills/_reference/endpoints.md).
• If a hub enables Hub Security, every hub HTTP/WS call needs a session cookie from POST /login first; the mechanisms must switch to the authenticated path.

Deploy is optimistically concurrent

• The version integer returned by /app/ajax/code and /driver/ajax/code is bumped on every save and is the concurrency stamp. An update must send the current version.
• If the hub rejects the version, a newer edit exists on the hub — re-pull and reconcile. Never blindly retry with an incremented number; that is how you clobber a change made in the web editor.
• The endpoints are undocumented and version-sensitive — re-verify after a platform update (skills/_reference/endpoints.md).
• The same version optimistic-concurrency pattern is not code-only. POST /device/update (device rename/edit) carries a per-device version stamp with the identical read-fresh-then-send rule — read it from fullJson immediately before each POST (skills/_reference/endpoints.md).