Heavy optimizations to the memory vector pruning. Agents now respond 3x faster in Slack integrations.