Long-Term Memory

Long-Term Memory (LTM) is the persistent knowledge layer that enables AI agents to retain and recall information across multiple sessions, moving beyond the stateless limitations of large language models (LLMs).

LTM is the critical component for building truly adaptive AI: it provides the capacity to store and retrieve data over extended periods, not just the limited context window (e.g., 128K tokens) of a single interaction. This is typically implemented using external storage systems like vector databases (for semantic search) and knowledge graphs (for relational context). For example, systems like Amazon Bedrock Agents or Google's Memory Bank use LTM to remember specific user preferences, past decisions, or accumulated domain knowledge. This enables continuous learning, personalization for recommendation engines (like Netflix or Amazon), and coherent, multi-session agentic workflows, transforming a transactional chatbot into a consistent, knowledgeable partner.

https://aws.amazon.com/blogs/database/build-persistent-memory-for-agentic-ai-applications-with-mem0-open-source-amazon-elasticache-for-valkey-and-amazon-neptune-analytics/