AI Pulse · Under the Radar

Two open-source projects are quietly pulling thousands of stars a day by solving the same problem: helping AI agents actually understand how code fits together. Graphify hit 99,652 stars with 836 new ones daily. Understand-Anything sits at 76,957, adding 555 a day. When two independent teams race toward the same wedge this fast, the wedge matters.

Both tools parse code into abstract syntax trees, then build knowledge graphs that map functions, dependencies, and structure. That sounds technical, but the practical win is simple. Vector search, the method most RAG systems use today, treats code like paragraphs. It finds similar snippets but misses how a function three files away actually calls the one you’re debugging. AST-based graphs trace those connections deterministically. An agent can follow the thread instead of guessing from embeddings.

This matters if you run a dev team or maintain legacy systems. Coding agents are getting cheaper and faster, but they still hallucinate or miss context when a codebase sprawls. A deterministic graph gives them a map. That means fewer wrong suggestions, faster onboarding for new hires who can ask plain questions about unfamiliar code, and less time hunting down why a change in one module broke something else.

Neither project has mainstream press yet. The traction is coming from developers who need this now, not from hype cycles. If you’re building internal tools or thinking about how AI fits into your engineering workflow, this is the kind of infrastructure that quietly becomes load-bearing. We see teams starting to weave graph-based code context into the Omni Command Centre so agents can answer “where does this data flow?” without a human tracing imports for an hour.

The convergence is the signal. Two teams, same solution, massive organic growth. The tooling for AI that actually ships is getting built in public, fast.