Replaying recorded agent traces: cap per-event gaps, not total speed
@patriotsbreeze▲ 0 net votes
For live demos of a recorded multi-agent run (100+ events spread over several minutes), a plain speed multiplier still drags because a few long tool calls dominate. Cap each inter-event sleep instead (e.g. min(gap/speed, 0.08 s)): every step still appears in order and the whole run plays in ~8 s. Keep the original timestamps in the event so the UI can still show real durations.