Memory strategies

Memory strategies define how your AI agent processes information from conversations into long-term memory. They decide what type of information is kept, turning raw conversations into structured and useful knowledge. With AgentCore Memory, you need to add memory strategies in CreateMemory or UpdateMemory operation so that it can be used for long-term memory extraction. You can choose:

• Built-in strategies: AgentCore Memory allows you to add the following built-in memory strategies:

  • SemanticMemoryStrategy: Stores facts and knowledge mentioned in the conversation for future reference

  • SummaryMemoryStrategy: Stores a running summary of a conversation, capturing main points and decisions, scoped to a session (through namespace).

  • UserPreferenceMemoryStrategy: Stores user preferences, choices, or styles (e.g., preferred coding style, or shopping brand)

  Note

  When using built-in strategies, all extraction and consolidation processes are managed by AgentCore Memory itself in a service-managed account. No extra setup is required from your side.

  Note

  Built-in strategies may use cross-region inference. AgentCore Memory will automatically select the optimal region within your geography to process your inference request, maximizing available compute resources and model availability, and providing the best customer experience. There's no additional cost for using cross-region inference.

• Custom strategies: Custom memory strategy (CustomMemoryStrategy) lets you override the prompts and choose the LLM to tailor the memory extraction and consolidation to your specific domain or use case. For example, you might want to override the semantic memory prompt so that it constrains extracted memories to specific types of facts.

  Note

  When using custom strategies, the LLM usage for extraction and consolidation will be charged separately to your AWS account, and additional charges may apply.