1 comment

[ 0.21 ms ] story [ 3.4 ms ] thread
Every time I open a new Claude Code or Cursor session on a project I've worked on for weeks, it starts from zero. It has to re-read files, re-grep the structure, re-guess what exists — every single session, forever. And when it tells me something about my codebase, I have no real way to check if it's true or just a plausible-sounding guess. That's what Rune fixes. It's a small runtime that sits underneath your AI coding tools, keeps a persistent "understanding graph" of your codebase up to date as you edit, and exposes it over MCP so any AI client can query it instead of re-deriving everything cold. The part I actually care about: every answer traces back to real evidence, not an assertion. Here's real output from my own repo tonight, asking for an architecture summary: Code $ rune explain derived_1 { "type": "architecture_summary", "description": "Detected stack: Express, React. 2 React component(s), 10 Express route(s) across the scanned source tree.", "confidence": "high", "evidenceChain": [ { "type": "react_component", "file": "components/UserCard.jsx", "line": 3, "evidence": "export function UserCard({ user }) {" }, { "type": "express_route", "method": "POST", "routePath": "/", "file": "routes/users.js", "line": 8, "evidence": "router.post(\"/\", (req, res) => {" }, ... 10 more, each with file + line + exact matched code ] }

It's not just claiming "this is an Express + React app" — it's showing exactly which facts, at which files and lines, prove it. Shipped an upgrade tonight moving the core extractor (imports, components, hooks) from regex pattern-matching to a real AST parser (Babel) — more accurate detection, fewer false positives. Route detection (Express/Next.js) is still regex for now — that's next, not hidden. Feedback welcome, especially on where it breaks.