JSON Mode

Agents pipeline LLM output into APIs, databases, and UI components. Invalid JSON breaks the chain; JSON mode prevents that class of failure.

JSON mode is a fill-in-the-blank form — the model must use the right format, not write a free-form essay you have to parse.

Scenario

An extraction agent returns {"entities": [{"name": "Acme", "type": "company"}], "sentiment": "positive"} — parsed directly into a database insert.

Solution

In Tool Calling & Function Calling, apply JSON Mode to this scenario: An extraction agent returns {"entities": [{"name": "Acme", "type": "company"}], "sentiment": "positive"} — parsed directly into a database insert. 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 JSON Mode (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 JSON Mode happens in the code.

1response = client.chat.completions.create(  # call the API2    model="gpt-4o-mini",  # key line for JSON Mode3    messages=[{"role": "user", "content": "Extract entities as JSON."}],  # key line for JSON Mode4    response_format={"type": "json_object"},  # key line for JSON Mode5)6data = json.loads(response.choices[0].message.content)  # key line for JSON Mode

• Treating JSON Mode as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for json mode

Quick recap — the most important points from this module.