Skip to main content

Documentation Index

Fetch the complete documentation index at: https://braintrust.dev/docs/llms.txt

Use this file to discover all available pages before exploring further.

Applies to:
  • Plan -
  • Deployment -
Summary To minimize costs while collecting user feedback for dataset curation, use init_logger() to create a logger instance and only call logger.log() when users provide feedback (thumbs up/down). This approach avoids automatic tracing of every interaction and allows manual review before promoting logs to datasets for experiments.

​
Configuration Steps

​
Step 1: Initialize logger at startup

Create a logger instance at application startup instead of using traced decorators for automatic logging. 
import braintrust

logger = braintrust.init_logger(project="My App")

​
Step 2: Log only on feedback events

Call logger.log() exclusively when user feedback is received to avoid logging every interaction. 
def on_feedback(user_input, agent_output, thumbs_up: bool):
    logger.log(
        input=user_input,
        output=agent_output,
        scores={"user_feedback": 1 if thumbs_up else 0},
        tags=["user-feedback"],
    )

​
Step 3: Review and promote logs

Review logged interactions in the project Logs tab, add additional tags as needed, and promote entries to datasets for running experiments.

​
Alternative: Direct Dataset Insertion

​
Skip logging layer entirely

For maximum cost savings, insert directly into datasets when feedback is received without using the logging layer. 
import braintrust

dataset = braintrust.init_dataset(project="My App", name="user-feedback")

def on_feedback(user_input, agent_output, thumbs_up: bool):
    dataset.insert(
        input=user_input,
        expected=agent_output,
        metadata={"feedback": "positive" if thumbs_up else "negative"},
    )