A support agent uses AgentCore Memory. It must interpret "that product" from the current conversation and reuse an explanation preference the user asked to save in a later conversation.
Warranty policies change periodically and must be retrieved from authorized official documents when answering. Do not assume that enabling memory automatically puts every saved record into the model input.
Which configuration best matches these three information roles?
Use short-term memory for raw conversational context, long-term memory for extracted preferences and insights across conversations, and RAG for managed business documents. Saving information does not automatically put all records into the model; retrieve what is needed and incorporate it into context.
Detailed explanation
Remembered conversational information alone does not establish the current contents of changing business documents.
A policy remembered from a conversation can become stale when official documents change. This scenario requires consulting current documents, so saved conversational information cannot replace the update check.
This separates immediate conversational context, persistent personal information, and changing business documents.
Use recent turns to resolve references, retrieve saved preferences under the storage policy, and retrieve current official documents. These roles provide both individual context and business evidence.
Repeated weight updates do not provide the required runtime conversational context in this scenario.
Fine-tuning adapts model weights through training. It differs from retrieving current conversation or individual saved information and does not replace the context-access requirement here.
Product documents have a different role from recent personal conversation and saved individual preferences.
Product retrieval fits the need for warranty evidence. Recent references and a preference the user asked to save come from different information sources and cannot be recovered solely from generic product documents.
Check official documentation
Compare the following points with the official documentation.
Read AIF-C01 Task 2.1.6, Memory types, and Compare long-term memory with RAG. Map raw events, extracted preferences, and current warranty documents to their roles.
Official references
- Official documentation 1
- Official documentation 2
- Official documentation 3
- Official documentation 4
- Official documentation 5
What to confirm
Distinguish short-term memory, long-term memory, and RAG; explain why short-term events need not vanish at session end and why saved data must be retrieved and incorporated into context.
Key points
- Events versus extracted records
- Memory versus RAG
- Storage versus retrieval
Notes
- Environment: Reference-based review of current AIF-C01 and AWS documentation using invented conversations and configurations; no real user-data storage, AWS API calls, or model inference.
- Check the distinction between raw events and extracted records in Memory types. Do not infer universal retention or deletion rules from the names short-term and long-term.
- Read the RAG comparison as a division of roles, not a guarantee that retrieval always ensures factual accuracy. Do not create memory resources, save real conversations, or run searches.
Foundation review
Short-term is not just a retention duration
AgentCore short-term memory handles raw conversation events. Retained events can restore context after a service restart or when continuing the same interaction later; the term does not mean they must disappear at session end.
Separate personal insights from changing business sources
Long-term memory can retain preferences and conversational insights, while RAG retrieves the current official policy. Both require attention to freshness, authorization, and purpose.
Connect storage to model input
Saving events, extracting records, and retrieving records are separate operations. The application retrieves needed information and incorporates it into context; this does not mean conversation automatically updates foundation-model weights.