Hey PH 👋 I'm Yagnesh, co-founder at AppUnik. I built Phasio after watching a prompt change silently break a customer-facing AI feature in production. No alert. No test. Nothing. Just users seeing garbage output. The problem: we unit test code religiously, but prompts? We were shipping them blind. Phasio fixes that. It's the Jest for LLM prompts — same describe/test API you already know, but for evaluating AI output. You write test cases, define validators (contains, notContains, llmJudge), and npx phasio runs them all. Exit code 1 on failure — drop it straight into GitHub Actions. Multi-provider from day one: run the same test suite against GPT-4o-mini and Claude Haiku simultaneously. Multi-judge scoring averages results across LLMs to reduce bias. It's open source (MIT). You bring your own API keys. No vendor lock-in. 👉 GitHub: https://github.com/YagneshKhamar... 👉 npm: npm install @phasio/sdk 👉 Docs + dashboard: https://phasio.dev Would love your feedback — especially from anyone who's been burned by a prompt regression in prod.

Forum Threads