# Concourse > Event-night agent for arena operators, built for the Qloo Agentic Hackathon 2026. It ranks a venue's own food, drink, merch and sponsor partners for the specific crowd coming tonight using Qloo taste data, lays the result out on a 3D arena, explains each pick and swaps stands on request. Data status: the hosted demo shows synthetic demo data shaped like Qloo responses until the hackathon API key is active. The UI labels the mode ("Demo data" or "Live data"). ## How it works - Qloo /v2/insights with filter.results.entities and feature.explainability scores the venue's shortlist against tonight's headliners (anchor entities resolved via /search). - Qloo urn:demographics gives the crowd's age and gender offsets (relative skews, not shares). - Qloo /v2/analysis/compare contrasts tonight's crowd with last night's. - A greedy assigner puts the best-fitting partner of each category on each stand, weighting premium partners toward club stands. - A Claude tool-use agent (tools: qloo_rank, qloo_demographics, qloo_compare, swap_stand) answers operator questions; without a model key a deterministic planner answers from the same plan. ## API (read-only, no auth) - GET /api/events: events, data mode, agent engine, featured event id - GET /api/plan?event=: the full plan (stand assignments, affinities, drivers, alternatives, demographics, compare diff, evidence ledger) - POST /api/agent {"question": "...", "event": ""}: agent answer, Qloo calls made, stand swaps ## MCP server - POST /api/mcp (JSON-RPC, MCP Streamable HTTP, stateless). Tools: list_events, plan_event, erlang_c, newsvendor, textbook_checks. - Plans the real-venue event-night decision for Climate Pledge Arena (flexible stand concepts, registers, par levels) and re-runs published textbook checks. ## Limits - Stand layout, vendor catalogue and event schedule are synthetic demo data. - Affinities are relative within one query; never compared across events. - "Per-cap lift" is a heuristic, not a Qloo output. - No personal data is sent to Qloo; nothing is inferred about individuals. ## Docs - README.md: setup, proof table, limitations - docs/qloo.md: enforcement layers, failure states, docs-vs-live differences - License: MIT