Things you can ask
Each one has its own system prompt, typed tools, and domain knowledge. The orchestrator picks who handles what.
Revenue trends, top products, customer segments, traffic sources, and anomaly detection.
Product descriptions, SEO optimization, pricing analysis, and bulk listing improvements.
Stock monitoring, restock recommendations, demand forecasting, and inventory updates.
Campaign planning, email copywriting, social media content, and discount strategies.
How the pieces fit together.
A central agent routes requests to specialized sub-agents using a flat tool map — no nested LLM calls.
Every tool queries a live Postgres database with 1,200+ orders, 200 customers, and 15k analytics events.
Complex requests are broken into workflows — the orchestrator plans, then executes each step sequentially.
Watch agents think in real-time — tool calls stream with animated step indicators and rich result cards.
Serverless-first. Neon for zero cold-start DB, Vercel AI SDK for streaming, server components where it makes sense.