LLMs are great proofreaders, and while rule-based spellcheckers exist, we noticed there wasn’t a good CLI tool that leveraged AI agents. So we built one! Spellcaster came together over a weekend at the Mintlify AI DevTools hackathon in San Francisco[0], where it won the Best Demo Award.
We ran Spellcaster against several large open source projects and identified hundreds of readability issues in their docs. Some popular ones:
* Cursor: 116 errors
* Mintlify 245
* Vercel: 1053 (My favorite was Vercel spelling "Lorem ipsum" wrong in one of their pages).
Spellcaster uses LiteLLM for easy model swapping and AgentOps for cost and error tracking.