SkeptAI

v0.2 update is officially complete! Shipped two features based on early feedback: 1. Optional API key field: paste your own Anthropic key to remove rate limits for your session. Never stored or logged. 2. Optional original prompt field: provide what you asked the AI for sharper Completeness and Contextual Fitness scoring in the CRIT report. Both features are live now at skeptai.dev. SDK coming in v0.3!

v.03 is coming soon!

@SmartMatchingjp Glad you're enjoying it! Feel free to offer any suggestions or constructive criticism!

@GetSkeptAI Been using this in production and the results are genuinely impressive. The key is knowing when to trust the AI and when to override.

v0.2 update is officially complete! Shipped two features based on early feedback: 1. Optional API key field: paste your own Anthropic key to remove rate limits for your session. Never stored or logged. 2. Optional original prompt field: provide what you asked the AI for sharper Completeness and Contextual Fitness scoring in the CRIT report. Both features are live now at skeptai.dev. SDK coming in v0.3!

Great news! You can now paste your original prompt for sharper Completeness scoring!

If you're enjoying using the platform and want to avoid rate limiting, you now have the ability to add your own API Key. Don't worry, your key will only be used for your browsing session until you close the window, and is never stored, logged, or transmitted anywhere except directly to the Anthropic API. Enjoy!

Don't forget that SkeptAI is open-source. We would love to see what the community builds. The roadmap has just been added to the github. Check it out: github.com/datonpope/skep…

We just launched on @ProductHunt ! 🚀🚀🚀 Go show us some love! producthunt.com/products/skept…

Sometimes, your LLM output needs to be reviewed by a "Digital Devil's Advocate" to spot the flaws. That's where SkeptAI comes in!

SkeptAI is free, open source, and no account is required. → Try it: skeptai.dev → GitHub: github.com/datonpope/skep… Hit "Try An Example" and watch an AI recommendation get destroyed in 45 seconds. I'm building in public. Follow me for the journey!

CRIT never uses the same model that produced the output. Claude output → critiqued by GPT-4o GPT-4o output → critiqued by Claude You can't get an honest critique from the same reasoning patterns that produced the original. Every report shows "CRITIQUED BY [model]" so you always know.

I pasted a Claude output into my own tool today. It got a 23/100. Disposition: REJECT This is why I built SkeptAI.