AI column

Research@donghyeok99Staff

TradingAgents: a stock report twelve LLM agents argue out

A trading blog ran it on NVDA: the bear researcher caught a supply-chain risk buried in the news, and the position came out smaller

Four analysts read NVDA at once, Bull and Bear weigh it on a scale, and one of five ratings lands in complete_report.html.
Four analysts read NVDA at once, Bull and Bear weigh it on a scale, and one of five ratings lands in complete_report.html.
Who
Tauric Research's research framework (arXiv 2412.20138), for developers and researchers who want a stock seen from several sides.
When
Opened in December 2024; the latest release is v0.7.1, from 27 October 2026.
Where
Your machine, Python 3.11+ or Docker; models via OpenAI, Claude, Gemini and other APIs, a ChatGPT Plus/Pro plan, or Ollama locally.

What was hard

Looking at one stock meant opening a chart, the news, the filings and the Reddit or StockTwits mood in separate tabs. Asking a chatbot once gave an answer, but no way to tell whether anyone had argued the other side.

What it does about it

TradingAgents copies a trading desk. Four analysts write their reports at the same time from their own data, a bull and a bear researcher debate them, and an aggressive, a conservative and a neutral risk analyst take apart the trader's proposal. The Portfolio Manager then reads lessons from how past calls actually paid off and picks one of five ratings.

The first screen: ticker, analysis date, LLM provider and research depth
The first screen: ticker, analysis date, LLM provider and research depth
Progress and reports fill in as the agents work
Progress and reports fill in as the agents work
A run from the README, at the trading decision
A run from the README, at the trading decision

What people made with it

Hacker Newslucemia512026-05-02

Rebuilt TradingAgents' team as a Claude Code plugin: /trading-analysis NVDA spawns seven subagents that run the analysts, the bull-bear debate and the managers on a Claude subscription.

I rebuilt it as a Claude Code plugin that runs on your existing Claude subscription — zero extra API cost.
likes 1 · comments 1 (2026-10-30 as of)Hacker News ↗
blog.pickmytrade.ioPickMyTrade2026-05-13

A run on NVDA in the Q1 2026 earnings season: bullish technicals, but the bear researcher flagged supply-chain exposure in recent news, so the final position came out smaller.

We ran TradingAgents on NVDA during Q1 2026 earnings season.

How it runs inside

  1. 1

    You → Analyst team (4)

    tradingagents --ticker NVDA --date …Pick a ticker, a date, the analysts and the models in the CLI, and the agent graph starts.cli/main.py
  2. 2

    Analyst team (4) → Bull & Bear researchers

    four reports: market, sentiment, news, fundamentalsThe analysts work in parallel on their own tools; the debate starts once every report is in.tradingagents/graph/setup.py
  3. 3

    Bull & Bear researchers → Trader & risk team

    investment plan with a five-tier ratingBull and bear trade arguments for the set number of rounds, and the Research Manager writes the plan.tradingagents/agents/managers/research_manager.py
  4. 4

    Trader & risk team → Portfolio Manager

    trade proposal and the risk debateThe Trader proposes timing and size; aggressive, conservative and neutral analysts argue over it.tradingagents/agents/trader/trader.py
  5. 5

    Portfolio Manager → complete_report.html

    Rating: Buy / Overweight / Hold / Underweight / SellWith lessons from past calls, it writes a rating line and its case, saved as section files and one HTML page.tradingagents/reporting.py

What goes in, what comes out

NVDA, 2026-09-01A decision on the five-tier scale and the same report tree the CLI saves
tradingagents backtest NVDA,AAPL --start 2026-06-01 --end 2026-08-01 --every 7Each decision on the ticker and date grid scored on alpha against its benchmark
tradingagents --portfolio my_book.jsonTrader and risk advice written against your own cash and positions

Is it for you

Good for

  • Getting one stock written up from several angles on a single page
  • Seeing how a multi-agent debate is wired in real code
  • Replaying the same process over dates and tickers and scoring its calls
  • Experimenting on local models through Ollama without an API bill

Not for

  • Per the README it is for research, not financial, investment or trading advice.
  • LLMs make the call, so the same ticker and date can come out differently from run to run.
  • Backtest results depend on the model, the date range and data quality, and are not guaranteed to match any published figure.
  • Runs on a ChatGPT plan count toward that plan's usage limits.

Reviewed by Devalley’s editor before publishing. How the column is made

Sign in to post, comment and like.Sign in

Report

Reason

Read the community rules