TradingAgents: Multi‑Agent LLM Financial Trading Framework
TradingAgents, open-sourced by Tauric Research, is a research-focused multi-agent trading framework built with LangGraph. Specialized agents (fundamental, sentiment, news, and technical) collaborate with trader and risk roles to evaluate the market and form decisions through a modular workflow. TradingAgents‑AI provides an accessible learning and hands-on entry built on top of this engine.
TradingAgents Engine Overview
A research and education oriented multi-agent trading engine centered on structured communication and dialectical collaboration. Analysts, researchers, trading, and risk roles work together to produce an explainable, evaluatable, and extensible decision process.
Architecture
Framework Structure
TradingAgents simulates a professional trading institution with clearly specialized roles: fundamental, sentiment, and technical analysts, researchers, a trader, and a risk management team. Roles collaborate through structured communication and debate to improve decision quality and optimize trading strategies.
- Analyst team: analysts collect relevant market information in parallel.
- Research team: researchers discuss and evaluate the collected data.
- Trader: makes trading decisions based on the researchers’ analysis.
- Risk management team: evaluates decisions and mitigates risk based on current conditions.
- Portfolio manager: approves and executes trades.
Goals & Key Properties
Integrate perspectives such as fundamentals, sentiment, news, and technical analysis into structured conclusions. Generate executable trading decisions under risk constraints, with order execution and logging in a simulated exchange.
- Modular: plug-and-play components for replacement and reuse.
- Explainable: reports and diagrams preserve key reasoning and evidence.
- Extensible: unified management for multiple models and data sources.
- Evaluatable: a closed loop via backtests, logs, and transaction records.
Role Specialization
Assigning dedicated roles to LLM agents helps decompose complex trading goals into manageable tasks. Inspired by real trading institutions, TradingAgents defines seven key roles: fundamental analyst, sentiment analyst, news analyst, technical analyst, researcher, trader, and risk manager. Each agent is equipped with role-aligned tools and constraints to ensure comprehensive market analysis and better-grounded decisions.
Analyst Team
Analysts collect and analyze market data from multiple angles:
- Fundamental analyst: evaluates company fundamentals and identifies potential mispricing.
- Sentiment analyst: analyzes social media and public sentiment to gauge market mood.
- News analyst: evaluates news and macro indicators to anticipate market changes.
- Technical analyst: uses technical indicators to forecast trends and opportunities.
These insights form the overall market view and are passed to the researcher team for further evaluation.
Research Team
Researchers rigorously evaluate analyst data through a dialectical process with bull and bear perspectives. Debate keeps analysis balanced, identifies opportunities and risks, and provides evidence for strategies.
Researchers
Bull and bear researchers conduct structured discussion and debate to balance return and risk, forming more robust views.
Trading & Risk
The trading agent consolidates analyses into decisions; the risk management team evaluates volatility and liquidity risks and advises the portfolio manager.
Risk Management
Risk evaluation and strategy adjustments; final approval by the portfolio manager, then order execution and logging in a simulated exchange.
- Bull researcher: emphasizes positive indicators and growth potential.
- Bear researcher: focuses on risks and negative market signals.
This dialectical process helps the trading agent make more grounded decisions with a balanced market understanding.
Trading Agent
The trading agent executes decisions based on comprehensive analysis. It evaluates analyst and researcher insights, chooses optimal actions, and balances return and risk in dynamic markets.
- Review analyst and researcher recommendations.
- Determine timing and position sizing.
- Execute buy/sell orders.
- Adjust the portfolio as the market changes.
- Precision and strategic thinking are key to performance.
Risk Management Team
The risk management team monitors market exposure and ensures trading stays within pre-defined limits.
- Assess volatility and liquidity.
- Apply risk mitigation strategies.
- Advise the trading agent on exposure.
- Align the portfolio with risk tolerance.
The team protects capital through effective risk control. All agents follow the ReAct prompting framework for collaborative and dynamic decision-making, closely resembling real trading systems.
Agent Workflow
Communication Protocol
To improve efficiency, TradingAgents uses a combined protocol of structured outputs and natural-language discussion. This approach preserves context across long interactions and reduces information loss, keeping communication focused and effective.
Interaction Types
Unlike older paradigms that rely heavily on unstructured dialogue, the system communicates primarily through structured reports and diagrams. Key information is retained and can be queried directly from global state.
- Analyst team: consolidates research into concise analytical reports.
- Trader: reviews reports and produces decision signals with detailed rationale.
Natural-language dialogue is used only when needed, such as debates between researchers and the risk team, to enable deeper reasoning and more balanced decisions.
- Research team: forms a balanced view through debate.
- Risk team: reviews trading plans across multiple risk dimensions.
- Portfolio manager: approves risk-adjusted decisions.
Backbone Models
Choose models by task: “fast-thinking” models for retrieval and utilities; “deep-thinking” models for analysis and decision-making. This balances efficiency and reasoning capability, runs without a GPU, and keeps the system adaptable to future model integrations.
Experiments & Results
Example Backtest
Cumulative returns curve for a sample asset. Results depend on model choice, time window, and data quality, and are for research demonstration only.
Transaction Log
Example execution trace to understand behavior, pacing, and risk exposure.
Framework & Platform
TradingAgents is an open-source, research-oriented multi-agent trading framework by Tauric Research, built with LangGraph and driven by role-based collaboration. TradingAgents‑CN provides a localized product and hands-on entry on top of it, using a FastAPI + Vue 3 architecture with enterprise capabilities such as auth, configuration center, and real-time notifications.
Multi‑Agent Engine
Fundamental, sentiment, news, and technical analysts, bull/bear researchers, trading and risk roles collaborate through a modular decision flow.
Localized Product Stack
FastAPI + Vue 3; MongoDB + Redis; auth & logs, configuration center, SSE/WebSocket real-time notifications.
Data & Model Management
Supports multiple LLM providers and model capability management; unified data sources such as Alpha Vantage, Tushare, AKShare, BaoStock, and more.
Features
Modern Architecture
FastAPI + Vue 3 for an enterprise-grade experience.
Database Stack
MongoDB + Redis for performance and multi-level caching.
Auth & Audit Logs
Authentication, role management, and operation audit logs.
Configuration Center
Visual configuration for models, data sources, and system settings.
Cache Management
Smart caching strategy with multi-level cache control.
Real-time Notifications
SSE + WebSocket for live progress and status updates.
Batch Analysis
Analyze multiple tickers concurrently for higher throughput.
Smart Screening
Multi-factor screening and ranking to support decisions.
Watchlists
Personal watchlists and groups for continuous tracking.
Ticker Details
Full ticker info and historical analysis records.
Simulated Trading
A sandbox exchange to validate strategies safely.
Dynamic Providers
Dynamically add and configure LLM providers.
Model Capability Profiles
Select and match models by task requirements.
Multi Data Sources
Unified management for Tushare/AKShare/BaoStock and more.
Report Export
Export professional reports to Markdown/Word/PDF.
Featured Videos
Releases
Original (English, Open Source)
TradingAgents
Maintained by the original authors. TradingAgents‑CN builds on top with localization and extensions.
Chinese Enhanced (Open Source)
TradingAgents‑CN v1.0.0-preview
A multi-agent quant framework for Chinese scenarios. Provides baseline analysis and an extensible architecture for learning and secondary development.
Hosted (TradingAgents‑AI)
TradingAgents‑AI v1.3.0-preview
Open registration, zero-deploy setup, with a small compute fee. Ideal for learning and trial use.
Pro (TradingAgents‑AI Pro)
TradingAgents‑AI Pro v1.0.0-preview
Pro features are under development. Registration requires an invite code.
CN Advanced Learner Edition
TradingAgents‑CN Pro
For learners who want to systematically improve trading skills: position analysis, trade review, and planning tools. Earn points via community tasks or tips to redeem access. This edition is for learning and simulation and does not provide direct buy/sell instructions.
Extraction code: g39k
Notice: The open-source projects and software referenced on this page remain the property of their original authors. We only provide technical consulting, deployment assistance, and community support under the applicable open-source licenses. Please comply with the corresponding licenses and terms of use.
Private Deployment & Configuration Service
We provide private deployment and configuration assistance for the open-source TradingAgents‑CN system, especially for users without a programming background. Via remote desktop, we help with installation, environment setup, and configuration. We also assist with creating LLM accounts (e.g., DeepSeek, OpenAI) and configuring market data sources. Service fee: 500 CNY per session. Contact the WeCom support below.
TradingAgents‑CN 1.0.0-preview
A localized product built on the TradingAgents engine with FastAPI + Vue 3. Supports multi-model and multi-data-source management, backtesting, simulation/live execution, and risk monitoring.
LLM Providers
| Platform | Overview | Sign‑up |
|---|---|---|
| OpenAI | GPT‑4 family (incl. GPT‑4o/4.1), o3; strong ecosystem | platform.openai.com |
| DeepSeek | Cost-effective models with strong reasoning/code and multilingual performance | platform.deepseek.com |
| Alibaba DashScope | Qwen models for text/vision/multimodal; cloud-native integration | dashscope.aliyun.com |
| Baidu ERNIE | ERNIE series optimized for Chinese scenarios; content/search integration | cloud.baidu.com/product/wenxin |
Data Sources
Tushare
Basic APIs are free; advanced APIs require points or membership
- Pros: comprehensive China A-share coverage, simple APIs, active community
- Cons: stability varies at higher frequency; advanced features depend on points
AKShare
Fully open-source and free, covering many data categories
- Pros: Python-friendly ecosystem; rich real-time and historical data
- Cons: depends on third-party websites; may break when sources change
BaoStock
Free historical quotes (China/HK/US)
- Pros: good data accuracy; stable minute/daily bars
- Cons: no real-time data; relatively limited feature set
Finnhub
Free tier is limited (5 req/min); paid tier increases limits and features
- Pros: global stocks/FX/crypto; real-time and historical; WebSocket
- Cons: strict free tier; access latency may vary by region
Yahoo Finance
Basic data is free; advanced APIs rely on third parties
- Pros: broad global coverage for non-commercial research
- Cons: API stability varies; historical gaps may exist
Alpha Vantage
Free tier (5 req/min, 500 req/day); paid tier increases quota
- Pros: rich technical indicators; well-documented API
- Cons: strict rate limits; some datasets update with delay
IEX Cloud
Free tier up to 50k requests/month; paid tiers for commercial use
- Pros: high-quality US equity data; low latency; rich dimensions
- Cons: mainly US-focused; other markets are limited
Wind
Module-based pricing; personal plans typically start at several thousand CNY/year
- Pros: comprehensive China-market coverage; authoritative data; strong analytics
- Cons: expensive; requires client software and additional API permissions
Eastmoney Choice
Personal plans around 3000–5000 CNY/year; institutional plans higher
- Pros: comprehensive China-market data; timely updates; quant APIs
- Cons: relatively expensive; client workflow can be complex; API requires application
Quandl
Basic free access; paid plans for professional/enterprise usage
- Pros: strong macro and alternative datasets for research
- Cons: limited equity market data; not ideal for high-frequency real-time
Community & Support
WeChat Official Account
Scan to follow for tutorials and updates.
WeCom Support
Scan to add support for consultations and integration help.