Let a Model Plan grep, Then Let Jev Rank the Code
BM25, dense embeddings, hybrid retrieval, and model-planned grep on 100 CodeNib Base issues: accuracy, latency, and cost across complete retrieval and reranking paths.
Read the articleCodeNib Blog
Implementation notes, measured tradeoffs, and explicit system boundaries behind CodeNib's repository analysis stack.
BM25, dense embeddings, hybrid retrieval, and model-planned grep on 100 CodeNib Base issues: accuracy, latency, and cost across complete retrieval and reranking paths.
Read the articleA reproducible local stack, a measured Qwen3.6 A3B versus Qwen3.8 plus DFlash2 comparison, and one verified manifest shared by the Wiki and coding agents.
Read the articleHow to connect CodeGraph to Codex or Claude Code, followed by its measured native query paths, provider-neutral facts, and explicit full-graph boundaries.
Read the article