AI Components
An AI Component is the unit you observe, evaluate, optimize, and compare. Choose the type based on the output you need—not the framework or model you happen to use today.
Prompt Optimization
Optimize a prompt for classification, text generation, or structured output. Compare candidates against repeatable examples and evaluators, then deploy the best-performing prompt.
Agent Optimization
Benchmark complete agents against versioned test cases and evaluators. Submit outputs, artifacts, and traces from your own runtime, then compare approaches on one leaderboard.
MCP Optimization Preview
Connect an MCP server, discover its tools, and improve their names and descriptions so a model selects the expected tool more reliably.
Tool Selection Optimization Preview
Paste an agent's tool definitions and optimize them against representative queries when an MCP connection is unavailable or unnecessary.
One component or several?
Create separate components when capabilities have different datasets, owners, deployment lifecycles, or success metrics. Keep versions of the same capability together so experiments remain comparable.
| Situation | Recommendation |
|---|---|
| One classifier with many prompt versions | One prompt component, multiple experiments |
| Support workflow and invoice extractor | Separate components |
| Same workflow in unrelated products with separate access | Separate AI Applications |
Trace attribution
Create a component before using its name for trace attribution. Trace ingestion never creates components implicitly.