RepoVault — AI analysis of 476 starred repos
A tool that takes all your starred GitHub repos and turns them from 'I'll need this someday' into a structured, searchable knowledge base — with an AI summary of each one.
The problem
GitHub stars are a graveyard of good intentions. You star something thinking “I’ll need this”, and 476 repos later you have no idea what you own. GitHub’s search doesn’t help — it searches by name, and you remember by idea.
The solution
RepoVault syncs your stars, walks each repo’s README and metadata, and asks the AI for a structured answer: what it does, what problem it solves, its architectural philosophy, when I’d use it, how mature it is, whether it’s Docker-ready. Ten fields, validated server-side, stored in PostgreSQL with pgvector for semantic search — you search by meaning, not by name.
Every AI request routes through my own AI Fusion gateway — RepoVault doesn’t even hold its own AI keys.
The meta moment
My favourite part: this tool’s output became the input for everything else. The export of all 476 analysed repos became the raw material for the improvement roadmap of every other project I run. The tool that organises other people’s knowledge ended up organising mine.