Company-wise memory in Amazon Bedrock with Amazon Neptune and Mem0

AWS has unveiled a new "company-wise memory" system for Amazon Bedrock, developed in collaboration with cybersecurity giant TrendMicro and memory platform Mem0. This architecture solves a critical enterprise AI problem: how to give chatbots persistent organizational knowledge while maintaining conversation context. The system integrates Amazon Neptune's knowledge graphs (storing company relationships and processes) with Mem0's memory management (handling both short-term conversational context and long-term persistent knowledge). Amazon Bedrock orchestrates the entire workflow, allowing AI agents to retrieve and apply this layered contextual information during inference.

TrendMicro implemented this system for their "Trend's Companion" chatbot to deliver personalized enterprise support. The technical pipeline begins with Claude models extracting entities and potential memories from conversations, which are then embedded using Amazon Bedrock's Titan Text Embedding model. These embeddings are searched against both Amazon OpenSearch Service (for semantic similarity) and Amazon Neptune (for structured knowledge graphs). A dual retrieval strategy ensures both flexible semantic matching and precise structured data access, with reranking models like Amazon Bedrock Rerank or Cohere Rerank prioritizing the most relevant information.

A crucial innovation is the human-in-the-loop feedback system. For each AI response, the system generates a "memory assessment report" showing which specific memories were referenced. Users can approve or reject these mappings—approved memories persist in the knowledge base while rejected ones are removed from both OpenSearch and Neptune. This continuous validation loop ensures memory accuracy and gives enterprises direct control over their AI's evolving knowledge, addressing trust and governance concerns that have hampered previous organizational memory implementations.