CodeNib Serves Repository Context to Coding Agents as a Reusable Multi-View Index, Cutting Trajectory Tokens 50 to 87% Versus grep and read
A fifteen-author system paper (arXiv 2607.25431, July 28) attacks the fact that coding agents rediscover the same repository structure on every task because indexes, language servers and task-local histories are disconnected. CodeNib builds reusable lexical, dense and structural views per repository commit, maps outputs to repository-relative source ranges, maintains selected views across edits, and serves ranked search, symbol navigation and bounded context through one runtime. Across 100 snapshots, graph and vector updates are 8.7x and 25.4x faster at the median than an independent rebuild, the median per-request live-server to static latency ratio is 4.7x on the 63% of 1,000 navigation requests that match normalized live locations, and across five models the selected-context policies preserve localization accuracy with 50 to 87% fewer trajectory tokens than paired grep/read.
↳ Follow the thread