Salesforce AI Research Releases VoiceAgentRAG: Dual-Agent Memory Router Cuts Voice RAG Retrieval Latency by 316x

Salesforce AI Research published VoiceAgentRAG, an open-source dual-agent memory router for voice RAG.

A foreground Fast Talker does sub-millisecond semantic cache lookups while a background Slow Thinker prefetches likely documents.

Benchmarks using Qdrant Cloud report a 316x retrieval speedup and a 75% overall cache hit rate across 200 queries.