面向Agent的一个代码搜索工具：Semble，用自然语言查代码库返回精准代码片段，比grep+read节省98% token消耗 方法是让Agent用自然语言直接定位到最相关的几行代码，不用瞎猜关键词，不用读整文件 速度上，索引一个普通仓库大约需要250毫秒，回答查询大约需要 1.5毫秒，NDCG@10=0.854 Semble用2k Token能到94%召回率，grep+read塞满…

Semble acts as a lightweight, super-fast search tool that allows AI agents to query codebases using natural language.

Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep (

Show HN: Semble – Code search for agents that uses 98% fewer tokens than grep (

spent way too long grepping agent code like a caveman. Semble just showed me a better way. 98% faster. tools > hustle. #buildinpublic

Code search for AI agents is quietly becoming its own infrastructure layer. Tools like Semble are replacing the “grep + open 12 files + waste 80k tokens” workflow with semantic retrieval that runs in…

Just discovered - it combines the key features from two YC startups: and This is a game-changer

minishlab/semble: Fast and Accurate Code Search for Agents. Uses ~98% fewer tokens than grep+read