AI21 Labs - Build Apps with Advanced AI Models

How does Maestro reduce agent costs without changing my code?

Maestro analyzes your running agents and applies automatic budget and compute scaling plus alternative execution paths that cut costs without modifying your application code. The framework identifies where your agent spends inefficiently (AI21 claims over 50% of typical agent budgets contain recoverable waste) and redirects work to cheaper options automatically.

What does in-flow validation do when an agent makes an error?

In-flow validation checks results at each step of your agent's execution and fixes errors automatically before they cascade through subsequent steps. This catches problems early in multi-step workflows rather than letting a single mistake corrupt the final output.

Can Maestro improve my RAG system's accuracy?

Maestro includes advanced RAG capabilities with structured data extraction and improved parsing from multiple document types, which can improve retrieval accuracy. The visual execution graphs let you trace where RAG results come from and debug retrieval problems.

How can I see which teams or agents are spending the most?

Maestro provides cost attribution broken down by team, repo, and individual agent, so you can identify where your spending goes. Combined with the visual execution graphs, you can trace expensive agent runs and understand what's driving costs.