Last released Oct 4, 2026
Evaluate and trace tool-calling LLM agents. Tool failure can never look like an empty result.