research / agent
TradingAgents
A multi-agent LLM framework for trading research with analyst, researcher, trader and risk management roles, run from a command line or Python.
At a glance
- Purpose
- A multi-agent LLM framework for trading research with analyst, researcher, trader and risk management roles, run from a command line or Python.
- Runs in
- Local agent with shell
- Operating systems
- Not stated
- Needs
- Runtime: Python 3.11+; Needs internet: Yes; Needs accounts or API keys beyond your agent: Not stated
- Files and data
- Reads files: Yes; Writes files: Yes; Sends data to other services: Yes
- Tool type
- Agent
- Delivery
- From GitHub
- Version
- Latest on the main branch
- Access
- Code on GitHub
- Compatibility
- According to the README, TradingAgents needs Python 3.11 or later. It shows conda and uv for creating the environment, and Docker as an alternative. It names LLM providers including OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, OpenRouter, Mistral, Kimi, Groq, NVIDIA, Azure OpenAI and AWS Bedrock, local models through Ollama, and OpenAI-compatible servers such as vLLM, LM Studio and llama.cpp. It says it works with any market Yahoo Finance covers, using the exchange-suffixed ticker.
- External costs
- The README names API keys for the LLM providers it supports, such as OPENAI_API_KEY, GOOGLE_API_KEY and ANTHROPIC_API_KEY. It also names a FRED key, which it calls free and optional, an Alpha Vantage key, and an optional TypeSafe key for Jev social-post screening. Azure OpenAI uses credentials in a separate .env.enterprise file, AWS Bedrock uses AWS credentials, and the README says local servers need no key. It says SEC EDGAR needs no account or API key. It states no prices, and the services you connect may charge for usage.
- Permissions
- The README shows installing the Python package and dependencies with pip or uv, or running it through Docker. The README requires credentials for the chosen hosted LLM provider and describes additional AWS credentials and Azure configuration. The README describes fetching market, news, filings and social data from sources including Yahoo Finance, SEC EDGAR, FRED, Alpha Vantage, StockTwits and Reddit. The README says completed runs append decisions to a memory log under ~/.tradingagents by default. The README says opt-in checkpoint resume stores per-ticker SQLite databases under ~/.tradingagents/cache/checkpoints, clears checkpoints on successful completion, and provides --clear-checkpoints to reset them before a run. The README describes saving reports and running analysis or decision settlement from scripts or scheduled jobs. The README says approved transaction proposals go to a simulated exchange.
- Usage terms
- Apache License 2.0, as stated in the repository. See its LICENSE file for the exact terms.
Code on GitHub
TauricResearch/TradingAgents
Apache-2.0TradingAgents: Multi-Agents LLM Financial Trading Framework
Topics: agent finance llm multiagent trading
Public GitHub data fetched · 0 days old
View on GitHub ↗What it does
According to the README, TradingAgents is a multi-agent trading framework with fundamentals, sentiment, news and technical analysts, bullish and bearish researchers, a trader, and risk management and portfolio manager roles. The README describes structured researcher debates and says an approved transaction proposal is sent to a simulated exchange.The README says the framework is for research and is not financial, investment or trading advice. The README describes an interactive CLI, non-interactive runs using flags and environment variables, and Python use through TradingAgentsGraph. The README also describes saved reports, a memory log, opt-in checkpoint resume, backtesting over tickers and dates, and passing a portfolio to a run.
The README says TradingAgents is built with LangGraph. The README lists providers including OpenAI, Google, Anthropic, xAI, DeepSeek, Qwen, GLM, MiniMax, OpenRouter, Mistral, Kimi, Groq and NVIDIA, alongside Azure OpenAI, AWS Bedrock, Ollama and OpenAI-compatible servers.
Who it is for
Researchers and developers who want to study multi-agent LLM analysis of financial markets.Example tasks
Install the package and launch the interactive CLI with the tradingagents command.Run an analysis without prompts by passing flags such as --ticker and --date.
Call TradingAgentsGraph from Python and read the decision that propagate returns.
Run a backtest over a grid of tickers and dates.
Pass your own portfolio so the trader, risk analysts and portfolio manager work against it.
Getting started
Clone the repository and create a Python 3.11 or later environment, for example with conda or uv. Install the package with:pip install .
Set the API key for your chosen provider as an environment variable, or copy .env.example to .env and fill in your keys. Then launch the CLI with either of:
tradingagents
python -m cli.main
For Docker, the README gives:
cp .env.example .env
docker compose run --rm tradingagents
The README says to add your API keys to .env first. It describes extra steps for Azure OpenAI, AWS Bedrock and local models.
The README has the exact commands.
Limitations
The README says TradingAgents needs Python 3.11 or later, provider credentials for hosted LLMs, and separate setup for Azure OpenAI and AWS Bedrock. The README says local model servers need no key.The README says the framework is for research and is not financial, investment or trading advice. The README says performance depends on models, temperature, trading periods, data quality and other non-deterministic factors, and that repeated ticker-and-date runs can differ. The README says backtest results are not guaranteed to match published figures. The README says social and news sources still reflect current content even for a historical trade date, and that reasoning models largely ignore temperature.
The README says historical runs withhold Yahoo company statements and insider trades. The README says EDGAR machine-readable filings begin in 2009 and fourth-quarter figures are reported as unavailable rather than derived from annual figures.
The repository can change at any time; check its README before you install.
Data practices
The README describes choosing an LLM provider, a local or remote Ollama endpoint, or an OpenAI-compatible endpoint. Prompt and portfolio transmission depends on the selected provider and endpoint; the README does not enumerate those payloads.The README says SEC EDGAR requests carry a default contact address and tells users to set their own with SEC_EDGAR_USER_AGENT. The README says setting TYPESAFE_API_KEY routes StockTwits and Reddit posts through TypeSafe's Jev for screening, while posts pass through unscreened without that key. The README also describes reaching Jev through OpenRouter.
The README says the memory log and opt-in checkpoints live under ~/.tradingagents by default. The README says the Docker setup stores results, reports, the memory log and cache in a volume. The README does not mention telemetry.