Platform Cookbooks

Add custom spans to any application

Connect spans across services with OpenTelemetry

Score every response as it is generated

End-to-end observability through a Portkey gateway

Query traces in natural language over MCP

Context-aware trace debugging with Falcon AI

Trace, debug, evaluate, and fix in one conversation

Write your own evaluation criteria

Evaluate at scale without blocking

Teach your evaluator what good means for your domain

Choose the metrics that drive optimization

Compare prompts and models on one dataset

Create test datasets from a column schema

Enrich rows with AI-generated data

Import datasets from Hugging Face

Human-in-the-loop annotation workflows

Curate regression datasets from production traces

Create, label, and serve prompt versions

Improve a prompt automatically

First steps with the agent-opt library

Optimize with evaluation in the loop

Evolutionary prompt optimization with GEPA

Improve prompts directly in your dataset

Bring your own data to agent-opt

ProTeGi, GEPA, and PromptWizard head to head

Bayesian search, GEPA, and meta-prompt compared

Cut LLM costs with semantic caching in Agent Command Center