Long-Term Memory

Short-term memory resets. Long-term memory makes agents feel personal and competent over weeks of interaction.

Long-term memory is a CRM record on every customer — the agent 'knows' you even if the current chat just started.

Scenario

After three sessions, the agent recalls you use Python 3.12, deploy on AWS, and dislike verbose logging — injected into the system prompt automatically.

Solution

In Agent Memory, apply Long-Term Memory to this scenario: After three sessions, the agent recalls you use Python 3. Identify the inputs, run the technique, validate the output, and note one thing you would monitor in production.

Do this before moving to the next module — reading alone is not enough.

Open the Code Walkthrough below and run it locally. Change one parameter related to Long-Term Memory (e.g. model, temperature, top_k, or tool name), observe the difference in output, and write 2–3 sentences explaining what changed.

Highlighted lines show where Long-Term Memory happens in the code.

1def recall_memories(user_id, query, vector_store, k=5):  # define a reusable function2    results = vector_store.similarity_search(  # key line for Long-Term Memory3        query,4        k=k,5        filter={"user_id": user_id},6    )7    return "\n".join(r.page_content for r in results)  # return the result

• Treating Long-Term Memory as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for long term memory

Quick recap — the most important points from this module.