0.27.1

Until now, every AI client that connected to a profile received its own private copy of that profile's tool catalogue and kept it for as long as the connection lived. For a profile carrying 286 tools that was several megabytes per client, and a busy workspace whose clients reconnect through the day was slowly filling the data plane's memory. That is gone. A profile's catalogue is now built once and shared by every client connected to it, so the data plane's footprint follows the number of profiles you run, not the number of AI clients you point at them.

Your edits reach the data plane within a second

Tick or untick a tool, connect or disconnect an app, attach a collection or publish a skill, and the shared catalogue for that workspace is refreshed within a second of the change being saved. Only that workspace's catalogues are touched; nobody else's is rebuilt because you changed something. The next client to connect to the profile sees the new tool list. A client that is already connected keeps the list it connected with until it reconnects, as before.

The analytics workspace is released on time

The temporary, in-memory analytics workspace that flc_load stages rows into has always been discarded ten minutes after its last use. The check for "ten minutes idle" used to run only when someone next used analytics, so a workspace left over from the end of one day could sit in memory until the first query of the next morning. It now runs on its own clock, every minute, whether anyone is querying or not.