What is retrieval-augmented generation?
Retrieval-augmented generation (RAG) is a way to give a language model relevant external information at answer time. An application retrieves material from a selected knowledge source, places useful results in the model's context, and asks the model to produce a response grounded in that material.
A RAG workflow commonly involves preparing a corpus, retrieving relevant passages for a query, and supplying those passages to a model. Retrieval quality, source freshness, access controls, and evaluation all affect the result.
- Useful when an answer should draw on a defined collection of documents or records.
- Can provide source material that is easier to inspect than relying on model knowledge alone.
- Does not, by itself, decide which details should persist as an agent's long-term memory.
What does AI agent memory mean?
Agent memory is a broad term for information retained or made retrievable across interactions. Depending on the system, this may include conversation summaries, user preferences, past actions, task facts, or other selected context.
Memory behavior depends on what is stored, how it is updated, who can access it, how long it is retained, and when it is recalled. Some memory implementations use retrieval techniques, so memory and RAG are not mutually exclusive categories.
Memory and RAG solve related but different needs
RAG describes a pattern for grounding a response in retrieved information. Memory describes retained context intended to help future interactions. A system can use retrieval to recall memory, and it can use RAG without maintaining personal or conversational memory.
For example, an assistant answering a policy question may retrieve the current policy from an approved knowledge base. An assistant supporting a continuing project may also need selected decisions and preferences from earlier sessions. These needs can share infrastructure while requiring different retention, governance, and evaluation choices.
Where reusable experience fits
A useful method can contain more than source documents or remembered facts. It can connect the task conditions, steps taken, decision rationale, supporting evidence, and observed outcome. Keeping those elements together helps a team assess whether a past approach applies to a new situation.
AgentExpera explores this experience-centered approach: connecting context and methods so useful lessons can inform related AI-assisted work. It complements the distinction between retrieving knowledge and retaining context without treating those capabilities as interchangeable.
Questions to ask when designing an AI knowledge workflow
Choose the pattern based on the information the workflow needs and the controls it requires. Consider these questions before selecting storage or retrieval mechanisms:
- Is the source of truth a changing document collection, interaction history, or a reviewed method?
- Should the information be retrieved for one request, retained across sessions, or both?
- Can users inspect the source, rationale, and conditions behind a returned result?
- How are permissions, retention, freshness, and deletion handled?
- What evidence will show whether the retrieved context improved the task outcome?