CryptoBot is a work-in-progress AI trading agent. It uses an LLM reasoning layer to interpret market data, manage risk rules, and decide whether to place, hold, or cancel orders. The goal is not to replace a trader, but to build a reliable, observable agent that augments decision-making.
What it does
- Market data ingestion: price, volume, and indicator streams.
- LLM reasoning: the agent receives a structured prompt with context and must decide the next action.
- Risk gates: position limits, stop-loss rules, and maximum daily trade counts prevent runaway behaviour.
- Exchange execution: order placement via REST/WebSocket APIs with confirmation and error handling.
- Audit trail: every decision is logged for later analysis and backtesting.
Why this matters for AI Test Engineering
Trading agents are excellent case studies for agent testing because mistakes have real consequences. A test harness here must cover:
- Tool-call correctness: does the agent call the right order endpoint with valid parameters?
- Reasoning stability: does the same market context produce consistent decisions across model versions?
- Safety constraints: does the agent respect risk limits under pressure and adversarial prompts?
- Latency: can the agent make and act on a decision before the market moves?
Current status
- Core reasoning loop: prototype
- Exchange API integration: in development
- Risk engine: defined, awaiting implementation
- Eval harness: planned
This project is experimental and educational. It is not financial advice.