Balance
204 posts

Balance
@0xBalanceXYZ
Web3 & AI growth studio. Currently building with @wise_node, @JudgeOnSol, @BitsoOnchain, and more.


While we gear up for the full Openhagen release, we're shipping some of the internal tools built along the way. First up is LazyLoadingAI; a structured retrieval layer for AI coding agents. As more engineers transition into agentic workflows, how well your agent understands your codebase matters more than ever. The problem is: they're terrible at it. Agents waste a ton of their token budget grep-ing and reading files they don't need. They explore endlessly, run out of context, and give you half an answer. LazyLoadingAI builds a local lightweight index of your codebase - symbols, call chains, type hierarchies and replaces blind grep/read loops with structured retrieval: search, hydrate, trace and 10 other tools. After comprehensive benchmarks on medium and large codebases: 48% fewer tokens · 41% lower cost · 1.7x faster Open source. Ships as an MCP server. npm install -g lazyloadingai cd your-project lazyloadingai init More details from @0xBoesgaard




The Official Marble Racing Game Built on @base

Had some extra time today which is great cause I got nerd sniped by kerneloptimization.fun - got it down to 1296 and can properly optimize down to low 1200s if I get more time. Great challenge.

Together with @0xBalanceXYZ I've been rethinking how AI agents navigate code as our current way of feeding them code feels flawed. 5,000 tokens → 200 tokens. Same context. Same quality. Can't wait to release it to all of you. More soon.






