Python Tool

LLMs are unreliable at arithmetic and data manipulation. A Python tool offloads precise computation to an interpreter the agent controls.

A Python tool is a calculator and spreadsheet rolled into one — the agent writes the formula, Python computes the answer.

Scenario

User uploads a CSV. The agent writes Python with pandas to compute monthly revenue trends and returns a matplotlib chart as base64.

Solution

In Tool Calling & Function Calling, apply Python Tool to this scenario: User uploads a CSV. 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 Python Tool (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 Python Tool happens in the code.

1def run_python(code, timeout=30):  # define a reusable function2    import subprocess, tempfile  # import dependencies3    with tempfile.NamedTemporaryFile(suffix=".py", mode="w") as f:4        f.write(code)5        f.flush()6        result = subprocess.run(7            ["python", f.name], capture_output=True, text=True, timeout=timeout  # key line for Python Tool8        )9    return {"stdout": result.stdout, "stderr": result.stderr, "exit_code": result.returncode}  # return the result

• Treating Python Tool as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for python tool

Quick recap — the most important points from this module.