⚡ The pipeline is fully streaming: audio segmentation, transcription and memory extraction run while the user is still speaking, with 0-300ms speculative prefetching 🎯 91.2% on #LoCoMo using only Top-5 memories, compared to 61.68% for #Mem0, and 69.44% on
@michabbb@social.vivaldi.net