Strategy-as-prompt trading agents built on AWS Strands + Bedrock AgentCore.
TradingStrands lets you describe a trading strategy in natural language — anything from "use turtle trading methodology with $1000" to multi-page theory-driven specs — and runs it as a live, risk-managed agent against real markets.
Strategies are compiled into a structured IR (threshold-to-action predicates + dynamic market-state observation schema) and executed by per-strategy agents under the supervision of a deterministic risk manager, modeled on a real-world trading desk.
There is no environment like production to test what happens in production.
No backtesting. No paper trading. Real money, real stakes, real markets — compensated by a what-if counterfactual tracker that records both trades taken and trades passed, marking them to market over time.
┌─────────────────┐
│ Orchestrator │ ticks every 1s (crypto) / 5s (equities, options)
│ (TTA check) │ wakes strategies on threshold crossings
└────────┬────────┘
│
┌────┴──────────────┬──────────────────┐
▼ ▼ ▼
┌─────────┐ ┌─────────┐ ┌─────────┐
│ Strategy│ │ Strategy│ ... │ Strategy│ each: Strands agent,
│ Bot A │ │ Bot B │ │ Bot N │ own ledger, own risk
└────┬────┘ └────┬────┘ └────┬────┘ tolerance, own TTA
│ intent │ intent │ intent
└──────────┬───────┴──────────────────┘
▼
┌───────────────┐
│ Trade │ normalizes & routes intents
│ Coordinator │ async dispatch to risk mgr
└───────┬───────┘
▼
┌───────────────┐
│ Risk Manager │ deterministic rules in hot path
│ (oversight) │ LLM advisory alongside
└───────┬───────┘ kill-switch authority
▼
┌───────────────┐
│Broker Adapter │ Robinhood (v0), Alpaca (v1)
└───────────────┘
- Equities — Robinhood (v0), Alpaca (v1)
- Options — same brokers; greeks are first-class in the IR
- Crypto — BTC / ETH only
Pre-alpha. See docs/SPEC.md for the authoritative design document.
MIT. See LICENSE.