{
  "id": 11424932,
  "title": "CodeSage: Code Intelligence for AI Agents, and Why I Rebuilt Its Models",
  "url": "https://urgent.news/2026/10/02/codesage-code-intelligence-for-ai-agents-and-why-i-rebuilt-its-models",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-02T11:23:47.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/iliaa/codesage-code-intelligence-for-ai-agents-and-why-i-rebuilt-its-models-2081"
  },
  "original_language": "en",
  "account": "CodeSage, a code intelligence engine for AI coding agents, has been rebuilt with new models and improved performance. Until recently, it utilized the Jina code embedder and ms-marco MiniLM reranker, but the full index of php-src took over half an hour to complete. In the newly rebuilt version (CodeSage 0.38), the same full index can be generated in just 373 seconds. The weights remained unchanged; the differences lie in the inference graph, the chunker, and a single SQLite query. The rebuilt models are now publicly available on Hugging Face, along with the scripts used to create them. CodeSage provides a structural graph (symbols, references, dependencies) and semantic search (embedding retrieval with cross-encoder reranking) in a single Rust binary, functioning as a CLI or MCP server. This allows AI agents to ask questions that grep cannot answer, such as identifying dependencies or locating specific code sections. The indexer updates automatically from git hooks on every commit, checkout, and merge, ensuring the index stays current. CodeSage's performance improvements were achieved by optimizing the inference graph, chunker, and SQLite queries, reducing the time needed to generate the full index from 1,782 seconds to just 373 seconds.",
  "summary": "Until this week, CodeSage ran its two models exactly as they ship on Hugging Face: Jina's code embedder and the ms-marco MiniLM reranker. A full index of php-src took 1,782 seconds, just under half an hour, and I got tired of waiting. CodeSage 0.38 does the same full index in 373 seconds. Nothing was retrained. The weights are the same weights; what changed is the inference graph around them, a…",
  "key_points": [
    "CodeSage rebuilt with new models and improved performance",
    "Full index of php-src now generates in 373 seconds",
    "Rebuilt models available on Hugging Face with CLI and MCP server"
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}