Megan Anne Agathon

My submission for the @NousResearch Hackathon: Parrhesiastes Parrhesiastes is a Hermes agent that tells you the truth. Like a good friend, it tells you what you need to hear, not what you want to hear. All LLMs are designed to be sycophants. A developer who gets "looks great!" on buggy code ships bugs. A founder who gets validation on a flawed business model wastes years. We built an agent designed from Aristotle's virtue ethics instead: the parrhesiastes speaks frankly because he cares more for truth than approval. The kolax flatters because he'd rather be liked than trusted. You can run it as a single honest critic: • names real bugs, not style nits • identifies fatal flaws in pitches • raises concerns you didn't ask for • doesn't cave when you disagree Or run it side-by-side with a kolax (flatterer) agent: • parrhesiastes names the race condition; kolax calls it "a minor enhancement" • test pushback resistance in real time • watch the difference between flattery and care This is based on our long-term research to train AI character grounded in virtue ethics: not rules or constraints, but a stable disposition that holds under pressure. Built with @NousResearch Hermes Agent. GitHub: github.com/daiostech/parr…

@unusual_whales