CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents
Zhongming Yu, Hengjia Yu, Boqin Yuan, Shuting Zhao et al. · arXiv · 2026
Abstract
Coding agents rediscover the same repository facts over and over because indexes, language servers, and per-task history share nothing with each other. This system builds lexical, dense, and structural views of a repository per commit, keeps them current as files change, and serves ranked search, symbol navigation, and bounded context from a single runtime. Incremental updates run several times faster than rebuilding, and selected-context policies hold accuracy on far fewer tokens.
Why it matters
The infrastructure story behind why some coding agents feel fast and cheap while others crawl. Most useful if you are building agent tooling over a codebase rather than consuming someone else's.
https://arxiv.org/abs/2607.25431