This example has a queue, prefill, reasoning, a tool wait, more reasoning and an answer. History retention is different from active model computation. The exact hosted sequence and cache management stay unknown.
This request example has visible output, model work, tool waits and retained history. Until the answer, the user may see only an optional preamble. A tool wait is different from active model computation. Longer work can keep state for a longer time, but the exact hosted sequence and retention stay unknown. The author made this explanatory diagram with example playback, and animation time is not measured runtime. Open full-size SVG