René's URL Explorer Experiment

    {
      "@context": "https://schema.org",
      "@type": "SoftwareApplication",
      "name": "code-review-graph",
      "applicationCategory": "DeveloperApplication",
      "operatingSystem": "Cross-platform",
      "softwareVersion": "2.3.4",
      "license": "https://opensource.org/licenses/MIT",
      "url": "https://github.com/tirth8205/code-review-graph",
      "author": { "@type": "Person", "name": "Tirth" },
      "description": "Local-first code intelligence graph for MCP and CLI. Builds a persistent map of your codebase so AI coding tools read only what matters, with 8.2x average context reduction in the current review benchmark and up to 49x in documented large-repo workflows.",
      "offers": { "@type": "Offer", "price": "0", "priceCurrency": "USD" }
    }
    

    {
      "@context": "https://schema.org",
      "@type": "FAQPage",
      "mainEntity": [
        { "@type": "Question", "name": "Are context savings exact?", "acceptedAnswer": { "@type": "Answer", "text": "No. Savings are estimated per response. They depend on the query, the repository, and which tools are invoked. We report estimates so you can reason about them, not as guarantees." } },
        { "@type": "Question", "name": "Will CRG always use fewer tokens than reading files directly?", "acceptedAnswer": { "@type": "Answer", "text": "No. For small single-file changes in tiny packages, graph metadata can exceed the raw file size. CRG pays off on multi-file changes, larger repositories, and architecture-level questions." } },
        { "@type": "Question", "name": "How accurate is impact analysis?", "acceptedAnswer": { "@type": "Answer", "text": "Impact analysis is conservative by design — it favours recall, so it can return false positives. In the current evaluation sample we see 100% recall and an average F1 of 0.54." } },
        { "@type": "Question", "name": "Does flow detection work for every language?", "acceptedAnswer": { "@type": "Answer", "text": "Flow detection is strongest in Python and framework-heavy cases today and continues to improve across other languages. Structural queries (call graphs, blast radius, communities) work broadly." } },
        { "@type": "Question", "name": "Do I have to use cloud embeddings?", "acceptedAnswer": { "@type": "Answer", "text": "No. Cloud embeddings (OpenAI, Google, MiniMax) are opt-in. Hybrid search works with local BM25 alone, and you can enable local sentence-transformers embeddings without sending code off your machine." } },
        { "@type": "Question", "name": "Is CRG only for Claude Code?", "acceptedAnswer": { "@type": "Answer", "text": "No. CRG is an MCP and CLI tool that works with any MCP-compatible AI coding tool, including Codex, Claude Code, Cursor, Windsurf, Zed, Continue, OpenCode, Antigravity, Gemini CLI, Qwen, Kiro, Qoder, GitHub Copilot, and GitHub Copilot CLI." } }
      ]
    }