MuhammadUsmanGM/CodeLens

CodeLens has a strong, useful premise: a CLI-launched Next.js app that indexes a repository, stores semantic chunks in Qdrant, and lets the developer chat with the codebase through Gemini-backed retrieval. The implementation is more than a landing-page prototype. Files like app/api/ingest/route.ts, lib/providers.ts, lib/ast-chunker.ts, lib/qdrant.ts, and lib/rag.ts show a real ingestion and retrieval pipeline, including multi-platform URL parsing, ZIP-based repo fetching, file filtering, SHA-256 file hashes for incremental indexing, provider-aware embeddings, full-context mode for smaller repos, vector retrieval for larger ones, and SSE streaming for chat. That gives the project a concrete technical spine.

The README is ambitious and gives a clear product story, with install instructions through npx @muhammadusmangm/codelens, a configuration table, technical limits, and a Mermaid architecture diagram. The .env.example is also helpful because it makes the Google and Qdrant requirements visible up front. I also like that the UI surfaces setup needs instead of letting users fail later, and that the project is MIT licensed.

The biggest adoption gap is trust and polish. There are no tests or CI workflows visible, which is risky for a tool that handles repo downloads, token-backed private repository access, chunking, vector writes, and LLM calls. I’d add focused tests around parseRepoUrl, provider fetch behavior, collection naming, file filtering, and retrieval mode selection first. The README and CLI output also appear to have encoding/mojibake issues around emoji and box-drawing characters, which makes an otherwise polished project feel less reliable. Finally, the changelog dates look inconsistent with the GitHub creation timeline, so I’d clean that up and add a short “privacy/security model” section explaining exactly what code is sent to Google, what stays local, and what Qdrant stores. That would make CodeLens much easier for cautious developers to try.