{"author":"tirthkanani","children":[{"author":"tirthkanani","children":[],"created_at":"2026-03-09T19:22:08.000Z","created_at_i":1773084128,"id":47314091,"options":[],"parent_id":47314090,"points":null,"story_id":47314090,"text":"Hi HN I'm Tirth. I built code-review-graph because I got tired of watching Claude Code re-read my entire codebase on every single task.

When you ask Claude Code to review a commit or add a feature, it reads files to understand the codebase. On a small project that's fine. On FastAPI (2,915 files) or Next.js (27,732 files) it scans thousands of files that have nothing to do with your change. You're paying for tokens that add zero value, and more noise makes the review worse.

code-review-graph builds a persistent structural map of your code using Tree-sitter. Every function, class, import, call, and inheritance relationship lives in a local SQLite database. When you edit a file or commit, it re-parses only the changed files and their dependants in under 2 seconds. Claude then queries the graph, finds what changed and what depends on it, and reads only the relevant files.

Benchmarks on production repos with real commits:

\u2022 httpx (125 files): 26.2x fewer tokens\n\u2022 FastAPI (2,915 files): 8.1x fewer tokens\n\u2022 Next.js (27,732 files): 6.0x fewer tokens on reviews, 49x on a live coding task\n\u2022 Review quality: 8.8 vs 7.2 out of 10

Some technical details:

\u2022 SQLite WAL mode for concurrent reads, no external DB\n\u2022 Qualified names (src/auth.py::AuthService.login) for collision-free node identity without scope resolution\n\u2022 SHA-256 hash skip: files touched but not modified are skipped entirely\n\u2022 Optional vector search stored as binary blobs in the same SQLite file, no separate vector DB\n\u2022 NetworkX for BFS graph traversal with a cached directed graph that rebuilds on writes\n\u2022 12 languages via Tree-sitter: Python, TypeScript, JavaScript, Go, Rust, Java, C#, Ruby, Kotlin, Swift, PHP, C/C++

No cloud. No telemetry. No sign-ups. One SQLite file in .code-review-graph/ and that's it. PostEdit and PostGit hooks keep the graph current automatically. Your workflow doesn't change.

Setup takes about 30 seconds:

  pip install code-review-graph\n  code-review-graph install\n
\nOr as a Claude Code plugin:\n claude plugin add tirth8205/code-review-graph

MIT licence. Around 3,700 lines of typed Python with 770 lines of tests.

GitHub: https://github.com/tirth8205/code-review-graph\n PyPI: https://pypi.org/project/code-review-graph/

Happy to answer questions about the incremental engine, the Tree-sitter integration, or the benchmark methodology.","title":null,"type":"comment","url":null},{"author":"chreniuc","children":[],"created_at":"2026-03-09T23:00:13.000Z","created_at_i":1773097213,"id":47316932,"options":[],"parent_id":47314090,"points":null,"story_id":47314090,"text":"I tried installing it using the recommended option, via Claude, it didn't work. I opened an issue and I've added what I have tried and what errors I did encounter.","title":null,"type":"comment","url":null}],"created_at":"2026-03-09T19:22:08.000Z","created_at_i":1773084128,"id":47314090,"options":[],"parent_id":null,"points":12,"story_id":47314090,"text":null,"title":"Code-review-graph: persistent code graph that cuts Claude Code token usage","type":"story","url":"https://github.com/tirth8205/code-review-graph"}