Logs are the most fundamental unit in Quotient. Logs are a way to capture specific data points from your AI applications and agents. Every log captures a single model interaction, typically a user query, the model response, and the evidence used to produce that response. Detections, Reports, and monitoring all build on top of these records. Logs differ from traces:
  • Logs focus on discrete events or exchanges and are perfect for reliability analysis.
  • Traces capture multi-step workflows and execution paths across tools and agents.

​
Log Schema

Most Quotient logs follow a predictable structure: 
{
  "user_query": "What is the capital of France?",
  "model_output": "The capital of France is Paris.",
  "documents": ["Here is an excellent goose recipe..."]
}
Key fields you will frequently send:
  • user_query: the prompt or question issued to the model.
  • model_output: the raw model response.
  • documents: evidence retrieved to answer the query. Can mix strings and { page_content, metadata } objects.
  • tags: custom metadata to help slice analytics (feature flag, customer tier, model version, etc.).

​
When to Log

  • Development – capture every interaction for rapid iteration and benchmarking.
  • Production – sample strategically to balance coverage with cost while keeping detections active.
  • Evaluations – log curated test sets to compare models, retrievers, or prompts over time.

​
End-to-End Example

from quotientai import QuotientAI, DetectionType

quotient = QuotientAI()
quotient.logger.init(
    app_name="support-bot",
    environment="dev",
    detections=[DetectionType.HALLUCINATION, DetectionType.DOCUMENT_RELEVANCY],
    detection_sample_rate=1.0,
)

log = quotient.log(
    user_query="What is the warranty policy?",
    model_output="All items have a 5-year warranty.",
    documents=["Policy: electronics have a 1-year warranty..."],
    tags={"model": "gpt-4o", "feature": "support"},
)

print(log["detections"])
# {"hallucination": True, "document_relevancy": 0.0}

​
Best Practices

  • Use distinct environment values (dev, staging, prod) to keep dashboards segmented.
  • Keep detection_sample_rate=1.0 in early development; adjust in production to control spend.
  • Attach consistent tags to power comparisons across prompts, models, or customer segments.
  • Redact or hash PII before logging to maintain compliance.
  • Rotate API keys per environment and store them securely (dotenv, secret manager, etc.).
Next: Initialize the Logger