Show HN: Proteus – Can DeepSeek Harness evolve itself to handle audio?

Category: ai-ml

Tags: self-improvement, agent-harness, llm-agents, evolution, benchmarking, research-preview

Score: 7.0/10 (Innovation: 8, Technical: 7, Documentation: 8, Utility: 5)

Proteus is a harness-agnostic framework for self-evolving LLM agent harnesses, letting you plug in any agent (DeepSeek, Pi, Codex, or your own) and measure how its prompt/memory/tools/skills change over many context-fresh episodes. It stands out by supporting goal, multi-goal, and no-goal regimes plus visible or hidden evaluators, and by shipping structural and behavioral measurement instruments (distance, crystallization, permutation tests) rather than just task pass-rates.

Target audience: AI/ML researchers and agent framework developers

Repository: https://proteus-evolve.github.io/dsh-self-improvement.html · Python · MIT · 109 stars

View on Hacker News