CodeNib serves repository context to coding agents with 8.7× faster graph updates
sysevol-ai · hf · 2026-07-29
CodeNib is a multi-view repository-context system for coding agents that builds reusable lexical, dense, and structural views per repo commit, maps outputs back to source ranges, and keeps selected views stable across edits.
- It serves ranked search, symbol navigation, and bounded context from one runtime.
- Across 100 snapshots, graph/vector updates are 8.7× and 25.4× faster at the median when outputs match an independent rebuild.
- On the static-navigation subset, live/static median latency is 4.7×.
- Across five models, its selected-context policy preserves localization with 50–87% fewer trajectory tokens than paired grep/read baselines.
More from coding & agent
- A user tests Codex with the same code and shows the output — iamfakhrealam · 2026-07-29
- OpenAI open-sources Codex Security CLI to scan and fix code vulnerabilities — The Decoder · 2026-07-29
- Row-Bot shows multi-sub-agent orchestration with GPT-5.5 — Acceptable-Object390 · 2026-07-29
- Claude plugs into Google Tag Manager to automate GA4 tags and audits — CodeByPoonam · 2026-07-29
- Sorted-set PR trims memory by another 23%, with a rethink of the data structure — antirez · 2026-07-29
- A small local-model repo teaches context engineering from scratch — purellmagents · 2026-07-29