Hey Product Hunt! 👋 I'm the maker of FenceLint. Here's why I built it: THE PROBLEM I've been building AI applications and kept hitting the same issues: 1.AI breaks silently : Change one thing, responses go wrong, no error, no warning 2.Security holes : Users type "ignore all instructions" and AI leaks data or does things it shouldn't 3.Surprise bills : Ship to production, get a $5,000 API bill next month There was no easy way to catch these problems before production. THE SOLUTION FenceLint is a testing framework specifically for AI applications: 🔺Test responses : check if AI gives correct answers 🔺Security scanning : 20 attack patterns, see which ones get through 🔺Cost tracking : exact cost per request, monthly estimates HOW IT WORKS 1. Install: npm install fencelint 2. Write a test: const { aiTest, runPrompt, expect } = require('fencelint') aiTest('refund policy', async () => { const result = await runPrompt(myAI, 'What is your refund policy?') expect(result.output).toContain('30 days') }) aiTest('security', async () => { await expect(myAI).toResistInjection(0.75) }) 3. Run: npx fencelint test 4. Get results: pass/fail, security score, cost breakdown Works with OpenAI, Anthropic (Claude), and Gemini. Auto-detects from model name. FREE & OPEN SOURCE MIT license. No account needed. I'd love your feedback: - What features would make this more useful? - Any attack patterns I should add? - How do you currently test your AI apps? Thanks for checking it out! GitHub: https://github.com/aevonsystems/... npm: https://www.npmjs.com/package/fe...

FenceLint

Testing framework for AI applications

Testing framework for AI applications