Turn any codebase into an explainable architecture map.
RepoWeaver is a local-first codebase intelligence cockpit. Give it a public GitHub repository or a local folder and it maps internal dependencies, identifies architectural hubs, traces change impact, surfaces evidence-backed risk signals, and generates a short onboarding tour.
No LLM key. No backend. No source upload.
The screenshot below is a live analysis of this public repository, Cosimo-Di-Rondo/RepoWeaver—not a mockup. RepoWeaver loaded the source from GitHub and detected 12 source files, 15 internal relationships, 93 symbols, and 5 architectural hotspots.
The selected App.tsx node shows its real inbound/outbound dependencies, complexity, symbols, and evidence-backed findings in the inspector.
Opening a new repository usually starts with the same archaeology: find the entry point, follow imports, guess which modules are central, and work out what a change might break. Existing code graph tools are powerful but often require a graph database, embeddings, or a model provider before they become useful.
RepoWeaver makes the first five minutes useful:
- Architecture map — internal file dependencies grouped by module and role.
- Impact tracing — reverse dependency traversal shows a change's downstream blast radius.
- Evidence-backed risks — each signal links to a file, line, and named rule.
- Generated code tour — entry point → architecture hub → risk boundary → safety net.
- Two import paths — public GitHub repositories and browser-local folders.
- Portable report — export the full analysis as JSON.
npm install
npm run devOpen http://127.0.0.1:4173. A built-in demo repository is ready immediately, or choose Analyze repo to load your own code.
npm test # static analysis and URL parsing tests
npm run build # type-check and production buildflowchart LR
A[GitHub URL or local folder] --> B[Browser source loader]
B --> C[Language-aware import scanner]
C --> D[Normalized dependency graph]
C --> E[Symbol and risk rules]
D --> F[Architecture map]
D --> G[Reverse impact traversal]
D --> H[Generated code tour]
E --> I[Evidence panel and JSON report]
The analysis engine is intentionally deterministic. It normalizes file paths, resolves relative JS/TS and Python imports, extracts common top-level symbols, scores branch complexity, and runs a small set of reviewable rules. The UI never invents an explanation: visible findings come from those rule results.
src/
├── components/GraphCanvas.tsx # deterministic graph layout and interactions
├── data/demo.ts # instant, self-contained product demo
├── lib/analyzer.ts # dependency, symbol, risk, tour, impact engine
├── lib/github.ts # GitHub and local-folder ingestion
├── App.tsx # product views and import workflow
└── styles.css # responsive visual system
- Useful before AI — architecture understanding should not depend on a paid model.
- Evidence over confidence — a line number beats a vague “AI detected a risk” claim.
- Progressive depth — overview first, then file evidence, then blast radius.
- Local-first by default — local source is read with browser file APIs and remains on-device.
- Demo in one click — the bundled repository exercises every major feature.
- Dependency resolution is strongest for JavaScript, TypeScript, JSX/TSX, and Python.
- Source inventory also recognizes Go, Rust, Java, CSS/SCSS, Vue, and Svelte files.
- GitHub's unauthenticated API limit applies to public-repository imports.
- The current graph is file-level; AST adapters and symbol-level call graphs are natural next steps.
See the implementation roadmap and the 2026-08-11 GitHub research notes.
RepoWeaver is an original implementation. Its product decisions were informed by patterns observed in successful open-source products—graph-native context, node-based exploration, one-step ingestion, auditability, and polished local tooling—but it does not copy their code or branding. The research notes document those influences explicitly.