EggMind

EggMind explores how agentic reasoning can make equality saturation more scalable and more directed. Instead of treating superoptimization as blind search, it runs a bounded optimization loop that proposes, tests, diagnoses, and revises rewrite strategies.

rewrite space
      |
      v
search with execution feedback
      |
      v
better optimization strategies

The high-level idea is to let execution evidence guide the search for useful rewrites. This makes the optimizer less like a brute-force enumerator and more like a system that learns which transformations are worth pursuing.