AgMoDB

Context for LLMs

AgMoDB (GSV-X4, "Agent/Model Database") is a tool in GSV's infrastructure domain that aggregates benchmark results and real-world evaluation data into a queryable format, organized by task class and cost point. Its purpose is to give agent pipelines an evidence base for model selection -- which model or agent to route a given task to -- rather than relying on intuition, brand preference, or a single isolated benchmark score. The mechanism is data aggregation at scale: 35 scrapers pull benchmark and evaluation data from external sources into a structured store, and a component called BenchPress produces predictions from that aggregated evidence. The stack is Next.js 16 on the frontend/API layer with Neon (serverless Postgres) as the database, and the product is live in production at agmodb.vercel.app rather than existing only as a prototype. AgMoDB sits at the intersection of three GSV clusters -- infrastructure, intercognition-jazz, and meaning-across-sapiences -- reflecting its role as shared plumbing that other agent-routing and multi-model-coordination projects in the portfolio can query rather than a standalone consumer product. Development is tracked under an ongoing epic (agmodb-ftw). A known, explicitly identified gap is the absence of formalized user personas: an April 2026 homepage-redesign brainstorm sketched five intended user-intent personas (building a product, personal use, research, batch use, and just exploring) to shape site information architecture, but none of these were ever implemented -- the live product today serves a single undifferentiated experience rather than persona-tailored views. Status: live and in-flight as of late May 2026, functional as a queryable benchmark/evaluation database with active scraping and prediction infrastructure, with the persona-driven IA layer as clearly scoped but unbuilt future work.