smolagents

• Key points are in the visual diagram above.

Some tasks are easier expressed as code than tool JSON. Code agents excel at math, data transforms, and multi-step logic with fewer round-trips than many single-purpose tools.

smolagents is giving the LLM a Python REPL instead of a Swiss Army knife — it writes the exact tool it needs on the fly.

Scenario

Agent receives a CSV analysis task, writes pandas code to compute cohort retention, executes it, and returns a chart path.

Solution

In Agent Framework Landscape, apply smolagents to this scenario: Agent receives a CSV analysis task, writes pandas code to compute cohort retention, executes it, and returns a chart path. 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 smolagents (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 smolagents happens in the code.

1from smolagents import CodeAgent, InferenceClientModel  # import dependencies2
3model = InferenceClientModel()4agent = CodeAgent(tools=[], model=model, additional_authorized_imports=["pandas"])5agent.run("Load sales.csv and compute monthly revenue trend.")

• Treating smolagents as a black box without evaluation
• Ignoring cost and latency in production
• Skipping error handling for smolagents

Quick recap — the most important points from this module.