Personal Agents and Conversational Memory

As AI agents become a primary interface for digital interaction, managing personal data acquired through conversations is increasingly important. This work focuses on conversational memory, the volatile knowledge that agents can accumulate about users over time, addressing challenges in structuring, updating, tracing provenance, and using such knowledge at query/inference time. We present preliminary work exploring structured representations to support the consolidation of conversational knowledge. As a proof of concept, we investigate the use of Knowledge Graphs to capture agentic conversational memory and propose a mechanism that combines imperative and generative computing to produce trustworthy and traceable responses. The approach enables transparent operation, traceable answers, and more deterministic responses grounded in explicitly stored information. A preliminary evaluation on a benchmark for long-term memory retention suggests the benefits of structured knowledge.