Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
WebMCP: SQL-First AI Memory
Live demo of an AI agent using WebMCP to run SQL queries on an in‑browser PostgreSQL database, navigating code graphs, issues, and workflow state.
Embeddings and semantic search were the first tools of choice to augment AIs with memory and help reduce hallucinations. Nowadays, however, most SOTA memory systems do not use embeddings or semantic search at all and instead let the AI explore a knowledge base with SQL.
I’ll demonstrate a live developer workflow where an AI agent uses WebMCP tools to query a PostgreSQL database running entirely in the browser (PG-Lite + WASM). You’ll see the AI’s SQL reasoning process in real-time as it explores project context, relationships between entities, and builds up understanding through structured queries rather than vector similarity.
WebMCP uses SQL-first, embedding-free AI memory for entity context management.
React/TypeScript WebSocket client tests Model Context Protocol servers using Drizzle ORM.
WebMCP exposes browser JavaScript functions as MCP tools using transports.