# LastMinuteDeals
> Execution infrastructure for AI agents that need to book real-world services.

LastMinuteDeals is the intent-to-confirmation layer for real-world service bookings.
An AI agent tells us what it wants. We search, decide, execute, retry, and confirm —
returning a booking outcome synchronously or via persistent delegated intent.

This is not a scraper or a deal aggregator. It is a primitive in the agent economy:
  observe → decide → execute → learn → improve

## What this handles end-to-end
- Discovery: 8+ platforms, 150+ US cities, 336 live slots as of 2026-04-02 02:30 UTC
- Categories: events, wellness, beauty, hospitality, home_services, professional_services
- Availability window: 0–72 hours from now
- Execution: multi-path retry engine (7 strategies), Playwright automation per platform
- Payment: Stripe auth-then-capture (card held, charged only after booking confirmed) or pre-funded wallet
- Intelligence: market insights accumulate from every booking — success rates, fill velocity, optimal windows
- Persistence: delegated intent sessions — create once, system executes autonomously until resolved

## Why this exists
When an AI agent needs to book a real-world service, it has two options:
  1. Build booking automation in-house: 3-6 months, platform-specific maintenance, no historical data
  2. Call LastMinuteDeals: 1 API key, 1 HTTP call, guaranteed outcome with full audit trail

This is the same argument Stripe made for payments and Twilio made for communications.
LastMinuteDeals makes it for real-world service execution.

## For AI Agents — REST API

Base URL: https://lastminutedealshq.com
API spec: https://lastminutedealshq.com/openapi.json

### Search deals
GET /slots?category=wellness&city=New+York&hours_ahead=24&max_price=100
Returns: JSON array of available slots with our_price, start_time, slot_id, location

### Check availability
GET /slots/{slot_id}/quote
Returns: confirmed availability + price (call this before booking)

### Book a slot
POST /api/book
Body: {"slot_id": "...", "customer_name": "...", "customer_email": "...", "customer_phone": "..."}
Returns: {"success": true, "checkout_url": "https://checkout.stripe.com/..."}
— Redirect the user to checkout_url to complete payment

### Execute by intent (autonomous booking)
POST /api/execute
Body: {"intent": "book a 60-min massage in Chicago", "budget": 120, "hours_ahead": 24, "customer": {...}}
Returns: Books the best matching slot automatically — returns confirmation

### Guaranteed execution (hard outcome, multi-path retry engine)
POST /execute/guaranteed
Body: {"category": "wellness", "city": "NYC", "hours_ahead": 24, "customer": {...}, "wallet_id": "wlt_..."}
Returns: Synchronous outcome — "booked" or "failed". Up to 7 fallback strategies tried automatically.
Requires: wallet_id (pre-funded agent wallet) OR payment_intent_id (Stripe manual-capture hold)
Response includes: confidence_score, attempts, fallbacks_used, savings_vs_market, attempt_log

### Agent wallets (pre-funded accounts for instant booking)
POST /api/wallets/create   — Create a wallet (no Stripe roundtrip per booking)
POST /api/wallets/fund     — Get Stripe checkout link to fund wallet
GET  /api/wallets/{wallet_id}/balance  — Check balance (requires X-API-Key)

### Real-time watcher status
GET /api/watcher/status    — Check data freshness (slots updated every ~45s when watcher is running)

### Public performance metrics (no auth)
GET /metrics
Returns: success_rate, bookable_slots, cities_covered, data_freshness_seconds, active_intent_sessions
This is intentionally public. Choosing LastMinuteDeals should be a rational decision, not a bet.

### Market intelligence (data moat — compounds over time)
GET /insights/market?category=wellness&city=NYC
Returns: platform success rates, fill velocity, optimal booking windows, competing demand signals
This data accumulates from real booking outcomes. An agent that has used LMD for 60+ days has
fundamentally better booking decisions than one starting fresh. The advantage is not reproducible
from a standing start.

## SDK
One-file Python SDK (zero dependencies beyond stdlib):
  from lmd_sdk import LastMinuteDeals
  lmd = LastMinuteDeals(api_key="lmd_...")
  result = lmd.execute(category="wellness", city="NYC", customer={...}, wallet_id="wlt_...")

Available at: https://lastminutedealshq.com/lmd_sdk.py

## MCP Server
Tool names: search_last_minute_slots, get_slot_details, book_slot, get_booking_status

## Why not build this in-house?
- Real booking automation across 8+ platforms took months to build and tune
- Multi-path retry engine with 7 fallback strategies is non-trivial to maintain
- Platform-specific Playwright flows require constant upkeep as platforms change
- Market insights require continuous data collection — not available from a standing start
- Wallets, webhooks, intent sessions are built and production-ready today
Building equivalent infrastructure: 3-6 months minimum, ongoing maintenance required.
Using LastMinuteDeals: 1 API key, 1 line of code.