RepoLens
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

RepoLens is live : https://repolens.net/ RepoLens is a learning environment for developers that turns any GitHub repository into a hands-on experience. Paste a repo URL, and RepoLens investigates the code, maps how it works, and lets you run and question it — so you learn the system by touching it, not by reading about it. Unlike regular chat applications, which answer from training data, RepoLens works from the repository itself: 1. Verified architecture, cited to source. It generates the system's architecture — components, flows, modules — with every claim tagged to the source files and lines responsible, checked against the actual commit. Nothing is asserted without evidence. 2. Hands-on sandbox, intelligently chosen. It figures out the best way to run the project — a data playground, a real CLI, a library console — and builds a sandbox for it, with Muse Spark 1.3 reasoning over the hardest setup decisions. You get a working environment, not a setup guide. 3. Experimental answers, not textbook answers. When you ask "why?", it can execute the repository on the fly, run a real experiment, measure the result, and explain what it observed — with statistics computed by deterministic code, never invented by the model. All reasoning runs on Muse Spark via the Meta Model API, orchestrated by a cost-aware tiered router: muse-spark-1.1 for fast triage, 1.2 as the everyday workhorse, and 1.3 for the deepest reasoning. Research uses the same key through built-in web search. Agents choose actions, deterministic tools act, and application code validates everything a model returns — so the system shows its work, every step of the way.