Working With Multi-Step Agents

AI Agents with Hugging Face smolagents

Adel Nehme

VP of AI Curriculum, DataCamp

Example: Travel Assistant

Planning Intervals: Helping Agents Rethink

agent = CodeAgent(
    tools=[document_search_tool],
    model=model,
    planning_interval=3,
    max_steps=12
)

planning_interval=3: Agent pauses after every 3 steps.

Agent Run with Planning Intervals

"Plan a 2-week family vacation to Europe with historical sites for adults and fun activities for kids, staying under $5000."

[Step 1] Search "Paris hotels" -> Found luxury hotels (~$4000 total)

[Step 2] Search "Paris attractions for families" -> Found Eiffel Tower, Louvre, theme park tickets

[Step 3] Pause and rethink (planning interval) Hotels are too expensive -> budget blown. I should look for cheaper options + kid-friendly activities.

[Step 4] Search "affordable family hotels in Europe" -> Found mid-range options in multiple cities

Callbacks: Hooks Into the Agent's Process

Basic Callback Function

def callback_function(agent_step, agent):
    # Do something with the agent step or the agent itself
    pass

agent_step: details about the step (plan, step number, etc.)
agent: the full agent object (state + methods)

Planning Step Callbacks

def planning_callback(agent_step, agent):
    print("AGENT PLANNING")
    print("=" * 50)
    print(agent_step.plan[:300])
    if len(agent_step.plan) > 300:
        print("\n... (plan truncated)")
    print("=" * 50)

AGENT PLANNING
==================================================
Search affordable hotels + kid activities
Then create balanced itinerary...
... (plan truncated)
==================================================

Action Step Callbacks

def action_callback(agent_step, agent):
    step_num = agent_step.step_number
    print(f"Step {step_num}: Taking action")

    if agent_step.is_final_answer:
        total_tokens = agent_step.token_usage.total_tokens
        print(f"Total tokens used: {total_tokens}")

Step 2: Taking action!
Step 3: Taking action!
Step 4: Taking action!
Total tokens used: 4,218

Adding Callbacks to Agents

from smolagents import ActionStep, PlanningStep

agent = CodeAgent(
    tools=[document_search_tool],
    model=model,
    step_callbacks={PlanningStep: planning_callback, ActionStep: action_callback}
)

What Can You Do with Callbacks?

Log searches to understand what users look for most
Add human-approval checkpoints
Send progress updates to dashboards or apps
Adjust agent behavior mid-run based on performance
And more...

Let's practice!

AI Agents with Hugging Face smolagents