Working Memory

The context window is the agent's RAM: finite, expensive, and shared across prompts, tools, and memory. Managing it is critical for long tasks.

Working memory is your desk surface — only what fits there is immediately usable; everything else is in a filing cabinet (long-term memory).

Scenario

During a 20-step coding task, the agent's working memory holds the current file contents, recent terminal output, and the last 5 user messages.

Solution

In Agent Memory, apply Working Memory to this scenario: During a 20-step coding task, the agent's working memory holds the current file contents, recent terminal output, and the last 5 user messages. 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 Working 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 Working Memory happens in the code.

1# Working Memory — minimal example2from openai import OpenAI3
4client = OpenAI()  # create API client5
6# Ask the model to explain this topic7response = client.chat.completions.create(  # core API call for Working Memory8    model="gpt-4o-mini",9    messages=[10        {"role": "system", "content": "You explain working memory clearly."},11        {"role": "user", "content": f"What is working memory?"},12    ],13    temperature=0,14)15print(response.choices[0].message.content)  # show output for debugging

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

Quick recap — the most important points from this module.