交易代理工作室 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 8.0 分,整体质量较高。我们强烈推荐将其纳入你的 AI 工具库,帮助提升工作效率。
交易代理工作室 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
交易代理工作室 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排,将复杂的多步骤任务拆解为清晰的自动化流程,实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成,适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一:pip 安装(推荐)
pip install tradingagents-studio
# 方式二:虚拟环境安装(推荐生产环境)
python -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install tradingagents-studio
# 方式三:从源码安装(获取最新功能)
git clone https://github.com/wjhccc/TradingAgents-Studio
cd TradingAgents-Studio
pip install -e .
# 验证安装
python -c "import tradingagents_studio; print('安装成功')"
# 命令行使用
tradingagents-studio --help
# 基本用法
tradingagents-studio input_file -o output_file
# Python 代码中调用
import tradingagents_studio
# 示例
result = tradingagents_studio.process("input")
print(result)
# tradingagents-studio 配置文件示例(config.yml) app: name: "tradingagents-studio" debug: false log_level: "INFO" # 运行时指定配置文件 tradingagents-studio --config config.yml # 或通过环境变量配置 export TRADINGAGENTS_STUDIO_API_KEY="your-key" export TRADINGAGENTS_STUDIO_OUTPUT_DIR="./output"
可视化多智能体 LLM 交易研究平台 — 看见 Agent 怎么想、怎么辩、怎么决策,而不是只看最后一个 BUY/SELL。 A visual multi-agent LLM trading research workbench. Watch the agents debate, see the causal chain unfold, replay history with one click.
English | 简体中文
⚠️ Research / educational tool only. Not investment advice. Full disclaimer ↓
---
---
```bash git clone <your-repo-url> TradingAgents-Studio cd TradingAgents-Studio
pip install -e ".[web,cn]" # Web UI + A-share (recommended)
```
from tradingagents.graph.trading_graph import TradingAgentsGraph
from tradingagents.default_config import DEFAULT_CONFIG
config = DEFAULT_CONFIG.copy()
config["llm_provider"] = "deepseek"
config["deep_think_llm"] = "deepseek-v4-pro"
config["quick_think_llm"] = "deepseek-v4-flash"
config["max_debate_rounds"] = 2
ta = TradingAgentsGraph(
selected_analysts=["market", "cn_social", "event", "news", "fundamentals"],
config=config,
)
_, decision = ta.propagate("贵州茅台", "2026-01-15") # or "600519"
print(decision)
A-share tickers are auto-routed through AKShare → Tushare → yfinance regardless of the global data_vendors setting. See tradingagents/default_config.py for the cn_data_vendors chain.
See examples/quickstart.py for a minimal runnable.
---
python -m venv .venv .venv\Scripts\activate # Windows
```bash cp .env.example .env
Studio bundles the muscles a research workbench actually needs:
| Feature | What it does |
|---|---|
| **Natural-language entry** | "研究茅台短期" / "AAPL 30 天" → ticker + date + period auto-filled. Rule-based first (deterministic, free), optional LLM fallback. |
| **Holdings tracking** | A-share / global positions with shares, cost, real-time quote, P&L, and **latest analysis signal per ticker**. CSV import accepts 代码/股数/成本价 Chinese headers. |
| **Scheduled analyses** | Interval / daily / weekly background runs. Analyst + LLM config snapshotted at create time. Auto-disable after 3 consecutive failures so a broken setup can't silently burn through your quota. |
| **Paper trading** | Virtual account, cash, positions, daily NAV snapshots. **One-click "按此决策模拟下单"** parses the trader proposal's Action + Entry/Target/Stop and opens a virtual position. Enforces A-share T+1. |
| **决策回放回测 (Decision Replay)** | Event-driven backtest replays Studio's stored Agent decisions over any window — answers *"if I'd followed the agents' Buy/Sell calls, what would my net worth look like?"*. **Zero LLM cost** since it replays history. Reports total return, max drawdown, Sharpe, Sortino, win rate, profit factor, alpha vs benchmark. Each trade links back to its source analysis report. |
| **决策质量看板 (Decision Quality)** | The next step after backtest. Scores **every individual completed analysis** against real N-day price moves (5 / 30 / 60-day horizons), benchmarked against the regional index. Surfaces overall win-rate / avg α / Sharpe, a **confidence-calibration curve** (does "0.8 confidence" actually win 80%?), breakdowns by **ticker / signal / single analyst / analyst combo / LLM model** (so you can answer *"did adding capital_flow improve alpha?"*), and a per-day calendar heatmap. Computed on demand — no extra tables, no LLM cost. |
| **K-line panel** | Per-ticker drawer from Holdings or Paper rows. Daily + 1/5/15/30/60-min bars, MA(5/10/20) + volume overlays, optional entry/target/stop reference lines, fullscreen mode. |
| **API key & model picker** | Per-provider model catalog (e.g. DeepSeek V4 Pro / V3.2 thinking / …, Claude Opus 4.7 / Sonnet 4.6 / …). API keys editable from Settings → written through to .env so the CLI sees the same values. Keys masked in read path, raw never echoed. |
Everything inherited from upstream — the LangGraph workflow, multi-provider LLMs, decision log, checkpoint resume — still works as before.
---
```
You can also manage LLM API keys from the Web Studio's Settings page — values are written through to .env so the CLI sees the same keys.
AI Skill Hub 为第三方内容聚合平台,本页面信息基于公开数据整理,不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后,再部署至生产环境,并做好必要的安全评估。
✅ Apache 2.0 — 宽松开源协议,可商用,需保留版权声明和 NOTICE 文件,含专利授权条款。
经综合评估,交易代理工作室 在Agent工作流赛道中表现稳健,质量优秀。如果你已有明确的使用需求,可以直接上手体验;如果还在评估阶段,建议对比同类工具后再做决策。
| 原始名称 | TradingAgents-Studio |
| Topics | AI交易LLM多智能体 |
| GitHub | https://github.com/wjhccc/TradingAgents-Studio |
| License | Apache-2.0 |
| 语言 | Python |
收录时间:2026-06-03 · 更新时间:2026-06-03 · License:Apache-2.0 · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型,复制安装指令后粘贴到对应客户端