AI

The site already shows Live, Historical, and Admin for humans. The AI layer is the same data, shaped for machines: stable JSON, tool schemas, and a project skill so Cursor (or any agent) knows the workflow without reverse-engineering the React UI.

1. Discover — GET /api/v1 then GET /api/v1/tools?format=openai.
2. Orient — get_current_event for the open 15m window (price to beat, odds, BTC).
3. Analyze history — list_history or POST /api/v1/query with filters (result, crosses, min/max Yes–No win %). Start with a modest limit; page with cursor.
4. Inspect a window — get_event + get_event_series for settlement and the BTC / win-% path inside that 15 minutes.
5. Use stored stats — get_stats (or aggregate=1) for SQLite aggregates already computed on this host.

Project skill at .cursor/skills/bitcoinbetting-api/SKILL.md tells the agent when to use the API and which endpoints to prefer. Keep that file updated if you add tools.

# Live context
"What is the current KXBTC15M window and distance from price to beat?"
→ get_current_event

# Screening
"Show the last 200 settled windows where Yes peak win % was at least 200, sorted high to low."
→ list_history / query with minYesWinPct + sort

# Deep dive
"For KXBTC15M-26OCT271930, summarize settlement and how often BTC crossed the beat."
→ get_event + get_event_series (+ crosses from history/stats)

• Large history pulls enrich crosses and win % from Kalshi — they can take time. Prefer paging over a single 10k call when possible.
• Peak win % can look extreme when the book briefly quotes a near-zero ask; treat outliers carefully.
• Set an API key in Admin (or EXTERNAL_API_KEY) to lock /api/v1/* behind a Bearer token on public deployments.

HTTP details: API. Tool catalog: Tool.