Cadenza connects WandB logs to research agents

Cadenza solves the data bottleneck for autonomous ML research by transforming flat logs into reasoning-ready snapshots.

• –Implements evolutionary concepts like "Islands" and "Elites" to filter and prioritize high-performing experiments for agent analysis
• –Provides a "sample" interface that generates compact JSON summaries optimized for standard LLM context windows
• –Enables a fully closed-loop research cycle where agents autonomously analyze past results, propose new runs, and update their own index
• –Requires a Luca AI API token, effectively serving as an integration layer for their broader Evolutionary Database ecosystem
• –Highly useful for researchers building "self-improving" training loops or autonomous agents that need to manage massive experiment histories