Show HN: Rendergit-lite – Interactive repo viewer with selectable files for LLMs (github.com)

Rendergit-lite is a fork of Andrej Karpathy’s rendergit that flattens any GitHub repo into a single interactive HTML page so you can pick exactly which files to feed an LLM. Installable via pip from the project repo (pip install git+https://github.com/peterdunson/rendergit-lite.git), a single CLI command (rendergit-lite https://github.com/user/repo) clones a repo, builds the page and opens it in your browser. The UI adds checkbox trees, quick filters ("Python only", "No tests", "Select all"), live selection stats, file-type icons, collapsible sections, mobile-friendly layout, syntax highlighting (Pygments), Markdown rendering, and an “LLM view” that emits CXML-ready snippets for ChatGPT/Claude and updates dynamically as you pick files. Smart defaults skip bloat (node_modules, package-lock.json, .venv), with flags to override (--keep-bloat), set file size limits (--max-bytes), save output (-o), or prevent auto-open (--no-open). For AI/ML practitioners, this is a practical tool for reducing hallucination and prompt bloat: instead of pasting an entire repo or noisy lockfiles, you can send only the relevant, real source files to an assistant, keeping context sizes manageable and improving reply fidelity. It’s useful for code review, debugging, dataset inspection, or targeted fine-tuning examples, and is permissively licensed (0BSD) for integration into tooling or pipelines.