tradingagents

tradingagents [--install-completion] [--show-completion] [--help] command [options]tradingagents analyze [--checkpoint | --no-checkpoint] [--clear-checkpoints]

tradingagents is the command-line entry point for TradingAgents, an open-source multi-agent framework that mimics a trading desk. Specialized LLM agents (market, sentiment, news, and fundamentals analysts; bull and bear researchers; a trader; risk analysts; and a portfolio manager) debate a ticker and emit a buy / overweight / hold / underweight / sell-style decision.The CLI is a Rich TUI. After the wizard, it streams agent status, tool calls, and report sections, then can save a report tree under `reports/` and display the full write-up. Results for each ticker and date are also written under ~/.tradingagents/logs/ (or TRADINGAGENTS_RESULTS_DIR). A decision log at ~/.tradingagents/memory/trading_memory.md records completed runs so later analyses can reflect on what happened.Tickers follow Yahoo Finance conventions, including exchange suffixes (`0700.HK`, `7203.T`, `AZN.L`) and crypto pairs (`BTC-USD`). The same package can be imported as a Python library (`TradingAgentsGraph.propagate()`).

analyze

Run the interactive CLI: pick a ticker, analysis date, output language, analyst team, research depth, LLM provider, and models, then stream the multi-agent debate.

Enable LangGraph checkpoint-resume for this run so a crash can continue from the last successful node. Omit the flag to honor TRADINGAGENTS_CHECKPOINT_ENABLED.

Disable checkpoint-resume for this run even if the environment enables it.

Delete all saved per-ticker checkpoint databases under the cache directory, then run analyze.

Install or print shell completion for tradingagents (Typer).

Show help for the program or a subcommand.

.env

Copied from `.env.example` in the project directory. Holds provider keys (`OPENAIAPIKEY`, `ANTHROPICAPIKEY`, `GOOGLEAPIKEY`, `XAIAPIKEY`, and others), optional FRED_API_KEY, OLLAMA_BASE_URL, and TRADINGAGENTS_* overrides.

Provider id (`openai`, `google`, `anthropic`, `ollama`, `openai_compatible`, ...). When set, the CLI skips the provider prompt.

Model ids for deep reasoning versus quick tasks.

Override the provider endpoint (vLLM, LM Studio, llama.cpp, or a remote Ollama).

Language for analyst reports and the final decision (internal debate stays in English).

Debate and risk-discussion round counts. Setting both skips the research-depth prompt.

Default for checkpoint-resume when --checkpoint / --no-checkpoint is omitted.

Checkpoint SQLite files live at `$TRADINGAGENTSCACHEDIR/checkpoints/<TICKER>.db` (default ~/.tradingagents/cache).

Decision-log path (default ~/.tradingagents/memory/trading_memory.md).

Per-run logs and reports (default ~/.tradingagents/logs).

The project is a research scaffold, not trading, investment, or financial advice. LLM sampling and live news/social feeds make two runs of the same ticker and date differ; reasoning models largely ignore temperature. Cloud providers and optional data vendors (Alpha Vantage, FRED) need API keys and incur usage cost. Requires Python 3.10+ (3.12 is the documented default). The interactive wizard needs a real terminal; piped or headless environments should set TRADINGAGENTS_* variables instead.

TradingAgents was introduced in 2024 by Yijia Xiao, Edward Sun, Di Luo, and Wei Wang (Tauric Research) in the paper *TradingAgents: Multi-Agents LLM Financial Trading Framework* (arXiv:2412.20138). The CLI, LangGraph pipeline, and provider registry have been developed in the open since then. Version 0.4.0 (August 2026) added working CLI checkpoint resume, look-ahead / point-in-time data fixes, and additional model coverage. Licensed under Apache License 2.0.

llm(1), ollama(1), freqtrade(1), pip(1)

Source code · Homepage · Documentation