What is TradingAgents?
TradingAgents, from Tauric Research, is a LangGraph framework that models a trading firm as a team of LLM agents. Fundamentals, sentiment, news and technical analysts work in parallel; bullish and bearish researchers debate their reports; a trader proposes a position; a risk team and a portfolio manager approve or reject it. It began as an arXiv paper (2412.20138, December 2024) and is labelled a research tool, not financial advice.
It runs as an interactive or scripted CLI and as a Python package (TradingAgentsGraph.propagate returns a decision), on any Yahoo Finance ticker from US stocks to Hong Kong, A-shares and crypto. Recent releases focus on look-ahead bias: SEC EDGAR fundamentals are served as filed on the run date, and a backtest grid scores past decisions on alpha from the first close after each decision.
The repository is Apache-2.0, has about 119,000 GitHub stars and 22,700 forks, and is at v0.7.1 (27 October 2026), with roughly one release a fortnight since September.
Is TradingAgents worth it?
TradingAgents earns its popularity less for its debate structure, which many imitators copied, than for the unglamorous work in recent releases on what the agents are allowed to know. A run dated in the past now reads SEC statements exactly as filed that day: the README's example is Apple's 2008 total assets, filed as $39.6B and restated to $36.2B in 2010, so a run in between sees $39.6B. Yahoo statements and insider trades, which cannot be dated by publication, are withheld from past-dated runs rather than leaked into them.
That matters because the backtest is the only way to tell whether the system decides well. run_backtest sweeps a grid of tickers and dates, enters each decision at the first close after its date, and scores realised alpha against a regional benchmark, keeping its own memory log so the live one is not contaminated. The README is frank that results will not match any published figure and vary between runs, because reasoning models ignore temperature and news sources reflect today.
Cost per decision is the thing to model before scaling. Every run calls up to four analysts, a multi-round debate, a trader, a risk debate and a manager, split between a quick and a deep model that can sit on different providers. Token use per run is not published. A ChatGPT Plus or Pro sign-in is supported and counts against the plan's limits, which suits experiments but not a nightly sweep of 500 tickers.
The framework is research scaffolding, not a trading system: it outputs a rating and rationale and sends approved orders only to a simulated exchange. With about 100 open issues and import paths that moved in v0.5.1, it rewards people who pin a version. Next to Vibe-Trading, it is narrower and easier to reason about: one ticker, one date, one debated answer.
Where it earns its place
A PhD student in computational finance
Compare LLM backends on the same trading decisions without look-ahead leakage
Run tradingagents backtest over a fixed ticker and date grid with each model in turn; scoring uses alpha from the first close after each decision. Budget for many model calls per cell, and resume interrupted sweeps with the same run_id.
A self-directed investor holding a dozen US stocks
Get a structured second opinion on a position before earnings
Run the CLI on the ticker with --portfolio pointing at a JSON of current holdings, so the trader and risk team reason about the actual book. Signing in with a ChatGPT Plus plan avoids API billing for occasional runs.
An engineer prototyping an internal research assistant at a wealth manager
Embed multi-analyst reports in an existing Python pipeline
Call TradingAgentsGraph.propagate and save_reports, with managers on Claude and analysts on a cheaper model via per-tier providers. Treat output as research input; the project says it is not investment advice.
Before you commit
- Setup
- Clone, create a Python 3.11+ environment, pip install ., set one provider key, then run tradingagents; Docker Compose and an Ollama profile are provided. A non-interactive run needs flags such as --ticker and --date.
- Integrations
- OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, OpenRouter, Mistral, Kimi, Groq, NVIDIA, Bedrock, Azure, Ollama and any OpenAI-compatible server; data from Yahoo Finance, SEC EDGAR, FRED, Alpha Vantage, StockTwits and Reddit.
- Data and privacy
- Reports, a decision memory log, cache and checkpoints sit under ~/.tradingagents. Prompts and fetched news go to the chosen LLM provider. The CLI checks for updates once a day without sending run data; TRADINGAGENTS_NO_UPDATE_CHECK=1 turns it off. Apache-2.0.
- Cost at scale
- No paid edition. Cost is LLM tokens per decision across analysts, two debates and managers, multiplied by tickers and dates in a backtest grid; per-run token counts are not published, so measure on a small grid first.
- Limits
- About 100 open issues and pull requests; non-deterministic output; Yahoo news feed outages have forced workarounds; EDGAR covers only SEC filers and reports fourth quarters as unavailable.
What’s good
- Point-in-time SEC fundamentals and withheld undatable data
- Backtest grid scored on alpha from the next close
- Quick and deep model tiers on separate providers
- More than a dozen LLM providers, plus local Ollama and vLLM
What’s not
- Many LLM calls per decision; token cost not published
- Runs are non-deterministic by design
- Simulated exchange only; no broker execution
- Import paths changed in v0.5.1; pin versions
Best for
Researchers and technical investors who want a well-documented multi-agent baseline for single-stock decisions and a fair way to backtest it.
TradingAgents alternatives
TradingAgents: common questions
- Is TradingAgents worth it?
- TradingAgents is the reference design for multi-agent stock analysis, and its recent work on look-ahead bias makes it one of the few you can backtest without fooling yourself. We rate it 4.0 out of 5.
- How much does TradingAgents cost?
- Free and open source (Apache-2.0). You pay your LLM provider or use a ChatGPT Plus or Pro plan; FRED and SEC EDGAR data are free, Alpha Vantage is optional.
- What are the downsides of TradingAgents?
- Many LLM calls per decision; token cost not published; runs are non-deterministic by design; simulated exchange only; no broker execution; import paths changed in v0.5.1; pin versions.
- Who is TradingAgents best for?
- Researchers and technical investors who want a well-documented multi-agent baseline for single-stock decisions and a fair way to backtest it.
- How much does TradingAgents cost for a team or in production?
- No paid edition. Cost is LLM tokens per decision across analysts, two debates and managers, multiplied by tickers and dates in a backtest grid; per-run token counts are not published, so measure on a small grid first.
- Is TradingAgents safe for company data?
- Reports, a decision memory log, cache and checkpoints sit under ~/.tradingagents. Prompts and fetched news go to the chosen LLM provider. The CLI checks for updates once a day without sending run data; TRADINGAGENTS_NO_UPDATE_CHECK=1 turns it off. Apache-2.0.
- What are the best TradingAgents alternatives?
- The closest alternatives we have reviewed are Vibe-Trading.
Answers checked Oct 31, 2026.
Make TradingAgents? Show this rating on your site
