theusefulnerd/shopkeeper-ai

ShopKeeper AI is a planned project that wraps a multi-agent AI system inside a retro pixel-art game. The README is upfront that the work is still in development: the design and architecture are done, and the build phase has not started yet. The idea is that a small shop is run by eight software agents who handle sales, inventory, payments, and fulfillment, while a human player can drop into the world from either side of the counter. As the customer, you wake up at home, open an in-game phone, chat with a Sales Agent, walk to the store, and check out. As the shopkeeper, you stand behind the counter and watch both walk-in customers and online orders on a live dashboard that shows what every agent is doing and why. You can swap roles mid-session, and whichever character you leave behind keeps running on its own. There are eight agents listed in a table. Some are powered by a large language model, including the Sales Agent that orchestrates conversations, a Recommendation Agent, a Supplier Agent that watches demand and suggests restocking, and a Post-Purchase Agent for returns and complaints. Others are plain deterministic code, including Inventory, Payment (a simulated UPI state machine), Fulfillment, and a Loyalty and Offers agent. The README notes that agents only talk to each other through a small structured state object, never through natural language. The planned stack uses LangGraph for agent orchestration, FastAPI with WebSockets for the API, Supabase Postgres for storage, Upstash Redis for cache, LangSmith and OpenTelemetry for tracing, Groq as the LLM provider, and a React plus Phaser 3 frontend deployed on Vercel and Render. The README does not state a license.