Attach metadata to traces to track experiments, environments, and system configurations. This metadata is essential for comparing performance across different setups and analyzing patterns in your AI agent behavior.

​
Usage

from atla_insights import configure

# Define metadata for tracking experiments
metadata = {
    "model": "gpt-4o",
    "prompt": "customer-support-v2",
    "experiment": "support-optimization"
}

# All subsequent traces will inherit this metadata
configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    metadata=metadata,
)

​
Key Metadata Tags

Focus on these three essential metadata tags for effective experiment tracking:
TagPurposeExample Values
modelTrack different models and versions"gpt-4o", "claude-3-sonnet", "gpt-3.5-turbo"
promptVersion control for prompt templates"baseline", "optimized-v2", "few-shot-v1"
experimentGroup related experimental runs"react-agent", "tool-calling-agent", "few-shot-agent"

​
Experiment Comparison Examples

​
Model Comparison

Compare error rates across different models: 
from atla_insights import configure

# Test with GPT-4
configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    metadata={
        "model": "gpt-4o",
        "prompt": "baseline",
    }
)
# Run your agent...

# Test with Claude
configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    metadata={
        "model": "claude-3-sonnet",
        "prompt": "baseline", 
    }
)
# Run your agent...

​
Prompt Optimization

Track different prompt versions: 
from atla_insights import configure

# Baseline prompt
configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    metadata={
        "model": "gpt-4o",
        "prompt": "baseline",
    }
)
# Run your agent...

# Optimized prompt with examples
configure(
    token="<MY_ATLA_INSIGHTS_TOKEN>",
    metadata={
        "model": "gpt-4o",
        "prompt": "few-shot-v1",
    }
)
# Run your agent...