Run multi-agent LLM financial trading analysis
A framework of multiple LLM agents that collaborate on financial trading decisions.
Done with a custom-built or unnamed agent
- What they did
- You install the repo, set an API key for an LLM provider, and pick a ticker and date in the CLI or through the Python TradingAgentsGraph call. Analyst agents cover fundamentals, sentiment, news and technicals. Bullish and bearish researchers then debate, a trader proposes a decision, and risk managers review it. The Portfolio Manager approves or rejects, and approved orders go to a simulated exchange.
- How it went
- The source reports no benchmark results. It says outputs vary between runs and backtest returns may not match published figures, so the framework is a research scaffold and not a strategy with a fixed return.
- Worth knowing
- Runs aren't reproducible, because LLM sampling and live news and social data change. Reasoning models ignore the temperature setting, so use a non-reasoning model for steadier output.
Try it yourself with Claude Code
Set up a small multi-agent research project in this folder where separate LLM agents act as a fundamentals analyst, a news analyst, a technical analyst and a risk manager for [ticker or list of tickers]. Have them debate and produce one written recommendation per ticker with reasoning, using only [free data source, e.g. yfinance]. This is paper analysis only, so never connect to a brokerage or place trades, and stop once you have run it on one ticker and saved the report.
Discussion on HN · Sep 8, 2026
Five of these in your inbox every morning
The best things people got an AI agent to do, each with the prompt to try it.
More like this
- Custom agent
Shop for umbrella insurance through your personal agent
Personal AI agents use Coverage Cat's Agent API/MCP to size umbrella coverage and compare carrier options with price ranges, backed by a licensed broker.
File a complex tax return with OpenClaw
OpenClaw built a Schedule C P&L from bank transactions, guided TurboTax entry for 4+ hours and caught an error in a prior accountant's work.
Fill forms and review tax filings with Codex computer use
Secondhand accounts of OpenAI staff using Codex computer use for forms and PDFs, including one tax filing review that caught an error worth thousands.
Prepare and file overdue German VAT statements
Cowork with Claude's Chrome extension prepared accounts from invoice PDFs, then logged into Elster and submitted the overdue Q3 2025 VAT forms.
- Custom agent
Draft a refund demand for an unauthorized renewal
The poster used ChatGPT to write a stronger message to Skillshare after an unauthorized $24 renewal, threatening a dispute and chargeback.
- Custom agent
Fight a mattress warranty claim by email
A user's agent emailed back and forth with the mattress company's agent, citing laws and pushing aggressively, to resolve a warranty claim.