[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fe31nyomgsph0":3},{"code":4,"message":5,"data":6,"timestamp":53},10000,"success",{"full_name":7,"position":8,"description":9,"language":10,"stars":11,"forks":12,"stars_gained":13,"age_days":14,"hn_threads":15,"one_liner":33,"summary_md":34,"topic":35,"github_topics":38,"license":43,"homepage":44,"updated_at":45,"history":46},"TauricResearch\u002FTradingAgents",7,"TradingAgents: Multi-Agents LLM Financial Trading Framework","Python",113716,22024,309,665,[16,22,28],{"id":17,"title":18,"points":19,"comments":20,"posted_at":21},49605822,"Multi-Agents LLM Financial Trading Framework",120,81,"2026-09-08",{"id":23,"title":24,"points":25,"comments":26,"posted_at":27},45189763,"CLI-based multi-agent trading system using LLMs",2,0,"2025-09-09",{"id":29,"title":30,"points":31,"comments":26,"posted_at":32},47954608,"Trading Agents – Multi-Agent LLM Financial Trading Network",1,"2026-04-29","A multi-agent LLM framework that debates and decides stock trades, for researchers and quant-curious developers.","## The problem\n\nMost LLM-in-finance attempts are a single prompt asking a model \"should I buy this stock?\" — one perspective, no accountability, and no way to trace where the reasoning came from. Real trading desks work differently: analysts, researchers, a trader and risk managers argue, and the disagreement is the product. TradingAgents encodes that org chart as LLM agents that research a ticker in parallel, debate bull and bear cases, then hand a proposal to risk management and a portfolio manager. It exists so you can inspect a structured, repeatable decision pipeline instead of a black-box answer.\n\n## What's different\n\nAgainst the typical approach — write a prompt, call an API, print a buy\u002Fsell — this is an opinionated role decomposition: fundamentals, sentiment, news and technical analysts run concurrently, a research debate pits bullish and bearish agents, and a separate risk team can veto the trader. That mirrors how discretionary funds actually allocate responsibility, and it gives you named intermediate artifacts (analyst reports, debate, risk assessment) you can audit.\n\nThe second differentiator is backtest integrity. The README and changelog repeatedly call out point-in-time and look-ahead fixes across FRED, social sentiment, SEC EDGAR filings and the decision-log memory, so a simulated run on a past date only sees what was published by then. That kind of hygiene is rare in hobby trading-LLM repos.\n\nThird, provider breadth: models from several vendors can be assigned per tier, so analysts, managers and the trader can run on different backbones.\n\n## Who it's for\n\nResearchers, students and quant-curious engineers who want to see how a multi-agent decision pipeline behaves end to end, and who are comfortable reading Python and experiment logs. Paper-adjacent work — the arXiv link and a technical report are here — suggests an academic origin, and the Hacker News thread drew serious discussion. It is also a reasonable scaffold for someone building a research prototype on top of the graph.\n\nIt is not for anyone who wants a bot that places real orders, and not for readers looking for a ready-made strategy with validated returns — the README explicitly frames this as research-only, with a disclaimer that performance varies with model choice, temperature and data quality. Traders without Python experience should look for a product, not a framework.\n\n## Cost of adoption\n\nPython 3.11+ (the docs suggest 3.13), plus API keys for an LLM provider and for data vendors — expect ongoing token and data spend, since each ticker runs several analysts plus a debate. Docker is offered as an alternative path, and reports, memory and cache go to a mounted volume. License is Apache-2.0, so commercial experimentation is unencumbered. The release cadence shown in the changelog is rapid — multiple releases per month, with import paths and data-provider integrations still moving, so expect occasional breakage between versions. The README is dense on architecture and thinner on measuring whether the decisions are any good.\n\n## Verdict\n\nTradingAgents is the rare case of a research repo that kept shipping: 665 days old, roughly 113.7k stars and 22k forks, still gaining (about +307 stars on the day it hit the trending board in October 2026). That is an established project with real mindshare, not a spike. Its value is the auditable pipeline and the point-in-time discipline, not a promise of alpha. Treat it as a lab bench: expect to spend on tokens, expect API drift, and judge it on the decision traces rather than on backtest numbers.",{"slug":36,"name":37},"finance","Finance",[39,36,40,41,42],"agent","llm","multiagent","trading","Apache-2.0","https:\u002F\u002Farxiv.org\u002Fpdf\u002F2412.20138","2026-10-24T23:06:01Z",[47,51],{"date":48,"stars":49,"stars_gained":50},"2026-10-24",113667,307,{"date":52,"stars":11,"stars_gained":13},"2026-10-25",1792898497]