Open-source engine

Self-host the research engine, or use the hosted desk

The research graph is public at github.com/tradingagents-report/tradingagents-report. Clone it, install with pip, and run the CLI. TradingAgents Report on this domain is the hosted desk around that engine: accounts, credits, and published reports.

Three objects, one name

TradingAgents is the academic framework (arXiv:2412.20138). The public engine at github.com/tradingagents-report/tradingagents-report is an Apache 2.0 derivative you can clone and run. TradingAgents Report on this domain is the hosted desk: signed-in jobs, usage-point billing, an admin catalog, and reports that stay private unless an administrator publishes them. If you only need the paper’s architecture, use the paper explainer. If you need to run a job here, use the method page.

Clone and run locally

Python 3.12, an LLM key, and optionally a TradingView Data API key. Redis and PostgreSQL are not required for a single-process run.

Without a TradingView key the engine falls back to Yahoo Finance and other enabled vendors.

Clone and install

Apache 2.0
git clone https://github.com/tradingagents-report/tradingagents-report.git
cd tradingagents-report
python3.12 -m venv .venv
source .venv/bin/activate
pip install .

Python 3.12. Redis and PostgreSQL are not required for a local run.

What this repository adds

The academic graph—specialist analysts, bull/bear debate, risk review, five-tier rating—stays in place. These are the product-side pieces that shipped with the public engine.

TradingView-first market data

Exchange-level symbols, daily bars, indicators, fundamentals, news, calendars, peers, and multi-timeframe TA. Yahoo Finance remains a fallback.

Structured decision cards

Analyst-section signals, a decision_brief, and in-window price and fundamental chart snapshots—not only role Markdown.

Multi-market ticker parsing

Deterministic listing identity for NASDAQ:AAPL, 600519.SS, HKEX:700, and other exchange-qualified forms.

In-process cache

TTL cache in process memory by default. Set REDIS_URL only if several processes analyze the same ticker and should share quota.

Three public sources

Paper, engine, and hosted desk

Three objects share a name. Keep them separate when you clone, cite, or run a job here.

TradingAgents paper

arXiv:2412.20138 and the Tauric Research repository. Original multi-agent design and the authors’ 2024 simulation. Not this product’s track record.

Public engine

github.com/tradingagents-report/tradingagents-report. Self-hosted CLI and Python runs with your keys. No accounts, billing, or public-report process.

Hosted desk

This site and agent.tradingagentsreport.com. Signed-in jobs, usage-point billing, admin catalogs, Simulated Trading, MCP, and curated public reports.

What the hosted product adds

Subscriptions, private analysis jobs, admin model catalogs, exchange whitelists, Simulated Trading, MCP, and curated public reports. None of that ships in the clone.

Research desk

Search listings, choose evidence modules, set analysis dates, and track jobs through to reports.

Usage-point billing

Plan and bonus points settle finished jobs from actual AI cost under product billing rules.

Curated public reports

Only administrator-published successful jobs appear on this marketing site. Desk runs stay private by default.

Install the hosted MCP →

What you can inspect in the repository

The public engine is the authority for the research graph you can run yourself. These are the parts that matter when you compare a local run with a report on this site.

The research graph

Specialist analysts, Bull and Bear debate, Research Manager, Trader, risk voices, and Portfolio Manager synthesis are encoded as a LangGraph you can read. A hosted job follows the deployed graph, which may omit or gate experiments.

Data and identity rules

Docs record look-ahead filters, listing identity, and vendor routing, with TradingView Data API first. Those rules exist so a historical analysis date does not quietly borrow later prices, filings, or news.

How people run it themselves

The project ships a CLI (`tradingagents-report`) and a Python entry point. You bring model and data credentials. Redis and PostgreSQL are optional. That local run is not this product’s billing, allowlist, or public-report process.

What this page does not claim

Thin origin pages often over-promise. These limits stay explicit.

  • Paper Table 1 and any local backtest are not this product’s track record.
  • A capability in the public README is not automatically enabled on the hosted desk.
  • The clone does not include accounts, usage-point billing, MCP, Simulated Trading, or the public-report library.
  • The hosted desk does not place live broker orders or hold customer assets.

Open-source FAQ

Questions that usually follow the repo link

Is the hosted product the same as the GitHub repository?

No. github.com/tradingagents-report/tradingagents-report is the self-hosted engine. This site is a hosted research desk with billing, private jobs, MCP, Simulated Trading, and curated public reports.

What do I need to run the engine locally?

Python 3.12, an LLM provider key, and optionally a TradingView Data API key from tradingviewapi.com. Redis and PostgreSQL are not required. Full multi-market coverage needs the TradingView key; without it the engine falls back to Yahoo Finance and other enabled vendors.

How does this relate to Tauric Research / TradingAgents?

The public engine is an Apache 2.0 derivative of github.com/TauricResearch/TradingAgents (arXiv:2412.20138). Cite the paper for the original design. Clone tradingagents-report/tradingagents-report to run this engine.

Does cloning give me billing, MCP, or Simulated Trading?

No. Those are hosted-desk features. The clone is the research graph, CLI, and data routing. Use this site’s method, MCP, and Simulated Trading pages for the product.

License

The public engine is Apache License 2.0. Keep the Tauric Research copyright notices in LICENSE and NOTICE. Third-party market-data and LLM vendor rights are not sublicensed.