CodeNib: A Multi-View Data System for Serving Repository Context to Coding Agents
CodeNib reports faster repository-context updates and lower token use by serving multiple reusable views of code at once.
The paper describes lexical, dense, and structural repository views built per commit, with outputs tied to source ranges. In its tests, graph and vector updates were 8.7x and 25.4x faster at the median when matching an independent rebuild. Static navigation was also faster than live-server locations on the matched subset, with a 4.7x median latency ratio. Across five models, selected context policies kept localization while using 50-87% fewer trajectory tokens than paired grep/read. HF Daily Papers' note
The paper describes lexical, dense, and structural repository views built per commit, with outputs tied to source ranges. In its tests, graph and vector updates were 8.7x and 25.4x faster at the median when matching an independent rebuild. Static navigation was also faster than live-server locations on the matched subset, with a 4.7x median latency ratio. Across five models, selected context policies kept localization while using 50-87% fewer trajectory tokens than paired grep/read. HF Daily Papers' note
score 5