Short-term, long-term and reasoning memory, rebuilt from the
agent_events rows that the ADK BigQuery Agent Analytics plugin wrote.
Every item points back to the row it came from.
Asked twice on the same day: to the agent with memory, and to the same agent without its memory tools. Both answers are the models' own, from real SQL on the same data.
Every session, by analyst and day. The arcs link the selected session to the earlier sessions its memory recall cited. Pick a session to open it.
Entities and facts extracted from this analyst's messages, and the preferences the agent saved, each from the row it came from.
One trace per turn: each model call, the tool calls it asked for, and how the turn ended. Hover a bar for its details.