Lepton Search
Creator · Lepton AI · 2024

You ask a question and get a written answer with numbered citations, the sources behind it, and some follow-up questions, in the format Perplexity made familiar. Lepton Search showed that this is mostly product design on top of commodity parts: a search backend, an LLM, and a stream. The whole thing is under 500 lines.
I created it at Lepton AI to show what could be deployed on the platform, and it went out as an open-source project in January 2024. It was one of the first open-source implementations of LLM search, and the most popular one at the time; Yangqing Jia, Lepton's co-founder, announced it on X. It uses whichever search backend you have a key for (Bing, Google, SearchApi, Serper), runs the results through an LLM, and caches each answer so a result has a shareable URL.
It reached over 8,000 stars. The repository is now archived, since this pattern has become standard.