Echolon
📖 English · 简体中文
An LLM-agent-native backtest framework for futures research. LLM agents driving quantitative research is becoming routine, and the trajectory looks inevitable. For agents to drive strategy creation reliably, they need backtest tooling designed for them — typed tools they call directly, structured error codes they can resolve, an indicator catalog they can query. Without that, agents writing trading code from prose documentation hallucinate: wrong indicator names, made-up function signatures, params that don't exist. Echolon makes the framework itself the agent's API: 23 MCP tools, 22 in-package skills, 32 catalogued error codes, 214 indicators with typed metadata. End-to-end today: SHFE daily futures.
Production engine inside Qorka, DolphinQuant's AI-native strategy generation product. Exercised by real money on SHFE every trading day.
Quickstart
Three commands cover the natural newcomer arc:
| Command | Purpose | Time |
|---|---|---|
echolon hello |
Quick demo. Downloads SHFE aluminum (last 2y) via akshare, scaffolds a strategy, runs a backtest. Network required. | ~30s |
echolon init <workspace> --market SHFE --instrument <i> --start <d> --end <d> --template <t> |
Start a real project. Downloads market data via akshare (free, no signup), scaffolds a strategy from a template, writes a workspace marker. | ~1–5 min |
echolon backtest single <strategy_dir> [--json] |
Iterate after editing. Walks up to recover ctx from the workspace marker, recomputes indicators, runs the backtest. No flags needed. | ~5–10s |
