Kraken meet AI - Try it

Place trades on Kraken with AI. Make trade orders in natural language (e.g. “Buy 1 ETH if a 10+ ETH sell prints”).

Evaluate live data against your condition and execute the order when matched.

View walk-through here: External link to Loom

Technical details

• Batch pipeline:
  • Parse incoming websocket messages, sort newest-first, then filter against the LRU in bufferTrades(parsed).
  • Accumulate accepted trades until the next rAF tick, then flush once to the store with scheduleFlush.
• Store flush:
  • Merge batch with existing trades and cap to MAX_TRADES_HISTORY.
  • Compute filteredTrades within filterRange and update only if IDs changed.
• Minimal re-render strategy:
  • Avoid notifying subscribers when visible data is unchanged, reducing React work under bursty traffic.
• Uses zustand:
  • Great for WebSocket-driven stores as there's no provider and minimal overhead.
  • Module-level refs buffer and coalesce updates between renders (e.g., pendingTrades, seenIds).
  • Stable subscriptions via use-sync-external-store under the hood.
• Deterministic natural language:
  • generateObject + zod makes outputs predictable. If there's no schema match, then no trade.