​
Overview

The update_step function allows you to modify an existing step’s information while it’s still active. This is useful for tracking progress, updating evaluations, or adding information discovered during execution.

​
Syntax

lai.update_step(
    step_id: Optional[str] = None,
    state: Optional[str] = None,
    action: Optional[str] = None,
    goal: Optional[str] = None,
    eval_score: Optional[float] = None,
    eval_description: Optional[str] = None,
    screenshot: Optional[str] = None,
    screenshot_path: Optional[str] = None
) -> None

​
Parameters

​
step_id
string
The ID of the step to update. If not provided, updates the current active step.
​
state
string
Updated description of the current state or context.
​
action
string
Updated description of what action the agent is taking.
​
goal
string
Updated description of what the agent aims to achieve.
​
eval_score
float
Updated evaluation score (typically 0.0 to 1.0).
​
eval_description
string
Updated text description of the evaluation.
​
screenshot
string
Base64-encoded screenshot to add or replace. Provide either this or screenshot_path.
​
screenshot_path
string
Path to a screenshot file. Provide either this or screenshot.

​
Returns

This function returns None.

​
Examples

​
Basic Progress Update

import lucidicai as lai

lai.init(session_name="Data Processing")

# Create initial step
lai.create_step(
    state="Starting data processing",
    action="Load and validate dataset",
    eval_score=0.0
)

# Update as processing progresses
for i, batch in enumerate(data_batches):
    process_batch(batch)
    progress = (i + 1) / len(data_batches)
    
    lai.update_step(
        eval_score=progress,
        eval_description=f"Processed {i+1}/{len(data_batches)} batches"
    )

lai.end_step()

​
Updating Based on Results

import lucidicai as lai

lai.init(session_name="Model Training")

step_id = lai.create_step(
    state="Model initialized",
    action="Train model on dataset",
    goal="Achieve >90% accuracy"
)

# Train model
model = train_model(data)
accuracy = evaluate_model(model)

# Update step based on results
if accuracy > 0.9:
    lai.update_step(
        step_id=step_id,
        eval_score=1.0,
        eval_description=f"Success! Achieved {accuracy:.2%} accuracy"
    )
else:
    lai.update_step(
        step_id=step_id,
        eval_score=accuracy,
        eval_description=f"Accuracy {accuracy:.2%} below target"
    )

lai.end_step()

​
Adding Screenshots During Execution

import lucidicai as lai

lai.init(session_name="UI Automation")

# Create step
lai.create_step(
    state="Browser opened",
    action="Navigate through checkout flow",
    goal="Complete purchase successfully"
)

# Navigate and capture screenshots at key points
navigate_to_cart()
lai.update_step(
    state="Cart page loaded",
    screenshot_path="/tmp/cart_screenshot.png"
)

fill_shipping_info()
lai.update_step(
    state="Shipping info completed",
    screenshot_path="/tmp/shipping_screenshot.png"
)

complete_payment()
lai.update_step(
    state="Payment processed",
    eval_score=1.0,
    eval_description="Checkout completed successfully",
    screenshot_path="/tmp/confirmation_screenshot.png"
)

lai.end_step()

​
Using with Decorators

import lucidicai as lai
from lucidicai.decorators import get_decorator_step

@lai.step(state="Initial state", action="Process data")
def process_with_updates(data: dict) -> dict:
    # Get the current step ID
    step_id = get_decorator_step()
    
    # Update as processing happens
    lai.update_step(
        step_id=step_id,
        state="Validation phase",
        eval_score=0.3
    )
    
    validated = validate_data(data)
    
    lai.update_step(
        step_id=step_id,
        state="Transformation phase",
        eval_score=0.7
    )
    
    result = transform_data(validated)
    
    # Final update
    lai.update_step(
        step_id=step_id,
        eval_score=1.0,
        eval_description="Processing completed successfully"
    )
    
    return result

​
Notes

  • Can only update active (not ended) steps
  • If step_id is not provided, updates the current active step
  • Updates are incremental - only provided fields are updated
  • Multiple updates can be made to the same step
  • Screenshot updates replace previous screenshots

​
Error Handling

from lucidicai.errors import InvalidOperationError

try:
    lai.update_step(eval_score=0.5)
except InvalidOperationError:
    print("No active step to update")

​
Best Practices

  1. Progress Tracking: Use eval_score to show progress (0.0 to 1.0)
  2. State Evolution: Update state description as context changes
  3. Meaningful Updates: Only update when there’s significant new information
  4. Final Evaluation: Always update eval_score before ending the step

​
See Also