LLM session
An LLM session is the grouped record of a multi-step interaction with a language model, linking the individual requests, responses, and tool calls of one task or conversation into a single traceable unit. It is the model-side counterpart of a browser session, and joining the two is where agent debugging gets its ground truth.
How it works
LLM observability tools group requests via session identifiers propagated through headers or metadata, producing a tree or timeline of the calls behind one agent run: prompts, completions, token counts, latencies, and tool invocations with their results. For agents that act on interfaces, the session trace explains intent while the interface recording explains effect; correlating them, usually by shared session id and timestamps, turns "the model chose to click submit" and "the page then showed a validation error" into one narrative. The rrweb custom event is a practical join point: dropping a marker into the DOM stream at each model call aligns the two records without a separate correlation store.
Where you encounter it
LLM observability platforms, agent debugging, and cost analysis, where sessions rather than single calls are the meaningful unit.