Hey PH! I'm Luis, a developer based in Germany. I built GSEP because prompt engineering doesn't scale. You write a great prompt today, it degrades tomorrow, and you're manually rewriting it for every user. GSEP applies biological evolution to fix this. Your agent's prompts are organized as 4 chromosome layers: - C0: Immutable identity (SHA-256 protected, never changes) - C1: Skills & tools (slow mutation, sandbox-tested) - C2: User preferences (fast, per-user adaptation) - C3: Content firewall (53 threat patterns) Mutations happen automatically. Fitness is measured across 6 dimensions. Bad changes roll back. Good ones propagate. The key insight: don't retrain the model — evolve the instructions. This works with any LLM without touching model weights. This is beta (v0.8.0). The evolution engine works and is tested, but we need real-world usage to validate the approach at scale. Your feedback genuinely shapes what we build next. npm install @pga-ai/core

GSEP - "Genomic Self-Evolving Prompt"

AI agents that evolve their prompts like DNA

AI agents that evolve their prompts like DNA