Memory Summarization

Summarization is the most practical compression technique — cheap with small models and effective for conversational memory.

Summarization is a journalist's lede — the whole story in two paragraphs so the reader can decide if they need details.

Scenario

After a 40-message planning session, summarization produces: 'Team chose React, deadline April 1, open question: auth provider (Auth0 vs Clerk).'

Solution

In Agent Memory, apply Memory Summarization to this scenario: After a 40-message planning session, summarization produces: 'Team chose React, deadline April 1, open question: auth provider (Auth0 vs Clerk). 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 Memory Summarization (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 Memory Summarization happens in the code.

1def summarize_history(messages, llm):  # define a reusable function2    transcript = "\n".join(f"{m['role']}: {m['content']}" for m in messages)3    prompt = f"Summarize this conversation. Keep facts, decisions, open questions:\n{transcript}"4    return llm.invoke(prompt).content  # return the result

• Treating Memory Summarization as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for memory summarization

Quick recap — the most important points from this module.