2.4. Sessions¶
The Sessions page (sidebar > Sessions) is a unified view of everything currently running in Backend.AI GO: the local inference servers behind your loaded models, and Squad agent loops. The Models tab manages model files on disk; Sessions shows their live runtime state.
Session Kinds¶
| Kind | Created when | Details shown |
|---|---|---|
| LLM serving | A model is loaded into memory | Engine, port, context length, memory usage, alias |
| Squad agent | A Squad agent loop starts | Squad, role, message count, last active |
Active Tab¶
The Active tab lists every running session with a live count, a refresh button, a state filter (All, Running, Starting, Busy, Idle, Error), and a search box.
Each session carries a state badge: Starting, Running, Idle, Busy, Stopping, Terminated, or Error. The Uptime value ticks up live while the session runs.
Card and Table Views¶
A toggle in the toolbar switches between card and table layouts. The table shows one row per session with columns for model/session name, status, engine, port, context, memory, uptime, and last active. Columns are sortable; sorting by uptime surfaces the longest-running sessions first.
Terminating a Session¶
Click Terminate on a session card (or row) to stop it. A confirmation dialog warns that active requests will fail and that the model or agent process shuts down immediately. Terminating an LLM serving session stops its inference server, which is the same as unloading the model.
History Tab¶
The History tab lists past sessions with their kind, duration, end time, and error (if the session ended abnormally). You can filter by kind (LLM serving / Squad agent), delete individual entries, and load older entries in pages.
Related Pages¶
- Running Models: loading models and the header status pill.
- Squad Planning & Execution: what Squad agents do while their session is active.