Alright, should be okay now. The types kind of threw me off. instanceId is treated as a string throughout the application with the undefined state being an empty string. There are some query wrappers that run when they should't but are disabled so long as "instanceId" is falsy. Should probably move to an explicit string | null type and restructure , but the exposure to that change across the app is huge.

For my own sanity, I'll try to explain the host behavior. In Rill developer, the host should be http://localhost:9009 when in dev mode. Instead of using the RILL_DEV flag in package.json, I've changed it to use the SvelteKit flag. In testing and prod, the host should be an empty string so that the requests are made via relative paths.

In web-admin, there's another httpClient for the admin service that uses Axios. It runs on http://localhost:8080 or, in production, on a host that is injected via an environment variable. This PR doesn't touch that.

web-admin also uses the runtime service, but the host is conditional based on the project and no calls should be made to it until you have a host and instance id. However, the layout does have hooks like useDashboards being called in the tree above the RuntimeProvider. Because of that, the web-common runtime store needs to be initialized with empty strings so that those queries are not enabled.