正在加载视频...
视频加载失败
over the weekend i built and shipped Sentra for our docs. we use Mintlify for documentation. the built-in assistant kept running into context issues. wrong answers, missing context, no citations. so we extended Sentra's capability to the docs site directly. now when you're reading Dodo Payments docs, you can... show more
18,408 次观看 • 2 个月前 •via X (Twitter)
13 条评论

The context and citation problem you hit is very real. Most of the built-in doc assistants are not great at answering with exact page citations. We’re building this space: connect your docs, github, etc and it handles customer onboarding and support with to-the-point, grounded answers. Our view is that you don't want companies pulled off their core craft to solve this, let us handle it. Give Revoca a try. We do a lot more than just surfacing the docs. I’d love to show you what we have.

@mintlify oof this is SO GOOD. Sire, the engineers g culture you’ve cultivated at Dodo is 🤌

@mintlify The citation layer is the real unlock for docs assistants. A wrong answer without a source feels worse than no assistant. I’d also make “I don’t know from the docs” a first-class path, with links to the closest source pages. That keeps trust intact when context is thin.

@mintlify I've been thinking of building this for myself any plans of making this open-source or as self-hostable Ask AI widget

@mintlify open sourced it here -

@mintlify Wow, thanks a lot Ayush! 🙌

@mintlify What LLM model you use under the hood?

@mintlify For this gpt-5.4-mini

@mintlify what's the reindex story when docs change, every deploy or on a schedule? that's the part that rots quietly, stale citations still read as grounded so you don't notice

@mintlify It reindex everyday to maintain freshness

@mintlify Actually i more like MCP because can conntect with my agent an AI

@mintlify This is also an extension to our mcp

@mintlify Nice direction. For docs assistants, citations are only the visible part. The deeper thing is trust recovery: show which page/version answered, what context was ignored, and where the answer is unsure. That makes wrong answers debuggable instead of just embarrassing.

