I built graphify after Karpathy’s /raw folder post. 6,000+ stars in 48 hours.

Inspired by AI pioneer Andrej Karpathy's April 2nd post about the potential for a 'raw folder' product, developer Safi Shamsi built Graphify, an open-source tool that rapidly gained over 6,000 GitHub stars in 48 hours. Graphify transforms any local folder—containing code, documents, papers, or images—into a persistent, queryable knowledge graph with a single command. Its unique two-phase processing begins with a completely deterministic, token-free pass using tree-sitter to parse 19 programming languages, ensuring no initial API calls or costs.

After the initial parse, Claude processes the content in parallel to build a rich graph where every relationship is tagged as 'found,' 'inferred,' or 'uncertain,' providing transparency. The graph persists across sessions, can be merged on updates, and automatically rebuilds with git hooks. Crucially, it operates entirely locally with no telemetry, vendor lock-in, or data leaving the machine, making it GDPR-safe by design. Early tests show dramatic efficiency; on a 6,100-file Unity codebase, it surfaced 3,957 hidden inheritance relationships while using 71.5x fewer tokens per query than reading raw files. It integrates natively with Claude Code via a simple install command, allowing an AI assistant to read the graph before every search.