Core Concepts

Understanding how FlautoPsy works will help you get the most out of drift detection.

Baselines

Learn how FlautoPsy learns normal behavior from your first 10 traces

Drift Detection Types

Understand the 5 types of drift that FlautoPsy detects

How It Works: Quick Overview

  1. You send traces (prompts, outputs, latency, cost) from your AI workflow
  2. After 10 traces, FlautoPsy learns your baseline — what "normal" looks like
  3. On new traces, FlautoPsy detects drift — changes in prompt, output, cost, latency, or topology
  4. You get alerts via Slack or in-app notifications

Why Baselines Matter

Without baselines, it's hard to know if a change is normal variation or a real problem. Baselines let us be smart about what's worth alerting on. For example:

  • If your workflow normally takes 500ms, a 3000ms call is worth alert
  • If your LLM usually outputs 200 tokens, a 5000-token response is unusual
  • If you've never changed your prompt, any drift in keywords is suspicious

Read more about how baselines are calculated.