virattt/ai-hedge-fund

5.4万星。 有人拿它搞量化，真有人赚了。 ai-hedge-fund，干什么的？ 模拟完整对冲基金团队。 分析师Agent、交易员Agent、风险经理Agent， 各角色有自己的目标、决策逻辑。 听起来很复杂对吧？ 但你知道这解决了多大的痛点吗？ 想入坑量化但不知道从哪开始？ 这个项目就是最佳起点。 底层接真实金融数据， 策略逻辑可落地。 有人做回测， 有人找灵感， 有人改改就实盘。 更狠的是，开源+可扩展， 你可以自己加策略、改Agent行为、接不同数据源。 我翻了一下代码， 结构清晰，文档齐全， 入门门槛比想象中低。 当然，风险提示： AI炒币有风险，入市需谨慎。 你对AI量化感兴趣吗？👇 github.com/virattt/ai-hedge-…

Matched through a repo-specific project phrase query.

又扫到了一个极其硬核的开源项目：ai-hedge-fund（已揽 5.6 万 Star）。 这就是个「多智能体 (Multi-Agent) 虚拟基金公司」。 它把华尔街大佬（巴菲特、木头姐、大空头 Michael Burry）和专业分析师（基本面、技术面、风控）全写成了独立的 AI Agent，直接帮你组团研判股票。 平时如果想搞一套带大模型的量化回测框架，光是接金融数据源、写并发协作逻辑、调教各个 Agent 的 Prompt 就能把人恶心死。（我深有感触） 这项目的精妙之处恰恰在于帮你跳过了前期造轮子的烂坑。它直接给了一套开箱即用的协作骨架，各路 Agent 各司其职交换完意见后，最后交由 Portfolio Manager 汇总拍板。 跑通成本极低。不用苦哈哈地配一堆环境，配好 key 后敲一行命令就能出分析和交易信号。 你想调 DeepSeek、GPT-4o 也行，想挂本地 Ollama 白嫖算力也行，连带历史数据回测（Backtester）和 Web 界面都给你备好了，把原来得折腾一两周的验证流程压缩到了几分钟。 想要仓库地址的同学举手👇👇👇👇

Matched through a repo-specific project phrase query.

Zoom? The AI hedge fund?

Matched through a repo-specific project phrase query.

Started building my personal AI Hedge Fund in the past few days. One summarized overview of all my bots and strategies, including logging, backtesting, and edge confirmation. The plan is to fully automate iterations and optimization with an agent hedge fund team using Hermes Agent.

Matched through a repo-specific project phrase query.

I ran every "ai hedge fund bot" posted in 2026 through a 6-month lookahead shift sharpe 8 backtests collapsed to sharpe 0.7 the models were memorizing prices from their training data the test is stupidly simple every bot's demo backtest sits inside the training window of its underlying llm gpt-4o cutoff → october 2023 every "sharpe 8 on aapl" screenshot → tested jan through march 2024 still inside so i shifted the window - pulled the exact prompt + agent stack from each repo - re-ran on a 6-month period after the model's cutoff - kept everything else identical results on 12 bots - 9 dropped from sharpe 4.8-8.2 down to sharpe 0.3-0.9 - 2 went negative - 1 actually held (it used only price ratios, zero news headlines) the paper calling this out dropped in january - arxiv 2601.13770 - benchmarks llm price recall - gpt-4o gets sp500 closes within 1% error before cutoff after cutoff the error goes up 40x the "alpha" is the model quoting its own training data back at you nobody runs the shift test not the authors not the people screenshotting equity curves not the funds tweeting they beat the baseline by 24% a separate arxiv paper (2505.07078) tested llm strategies across 100+ symbols over 20 years llms came out "overly conservative in bull markets, overly aggressive in bear markets" classic memorization artifact - the model remembers regime shifts, not signal takes 20 minutes to run this on any new bot i check every repo before i copy a single line the test is free the fraud is not

Matched through a repo-specific project phrase query.

The Rallies AI Hedge Fund just bought some Adobe $ADBE stock

Matched through a repo-specific project phrase query.

🚨 What if Warren Buffett, Michael Burry, and Cathie Wood all analyzed the same stock at the same time? Someone built exactly that. 12 AI agents. One verdict. Free. It's called AI Hedge Fund. 56,000 GitHub stars. The investing world has never seen anything like this. Not a dashboard. Not a screener. A full AI investment committee that debates your stock picks the way real hedge fund teams do — then gives you a buy, sell, or hold decision with full reasoning from every agent. Here's who's on your team: → Warren Buffett — only buys wonderful businesses at fair prices → Charlie Munger — demands a margin of safety on every pick → Michael Burry — The Big Short contrarian hunting deep value → Cathie Wood — innovation and disruption, high conviction growth → Bill Ackman — activist investor, takes bold positions → Ben Graham — the godfather of value investing, hidden gems only → Aswath Damodaran — the Dean of Valuation, story meets numbers → Plus a Technicals Agent, Sentiment Agent, Fundamentals Agent, Risk Manager, and Portfolio Manager Here's how it works: → Enter any stock ticker — AAPL, NVDA, TSLA, anything → Each agent pulls real financial data — earnings, balance sheets, insider trades, news, sentiment → Every agent analyzes the stock through their own investment philosophy → The Risk Manager checks position sizing and portfolio exposure → The Portfolio Manager takes all signals and makes the final call → You get a buy/sell/hold decision with full reasoning from every agent Turn on --show-reasoning and watch them think. Buffett breaks down the moat. Burry flags what everyone else missed. Cathie Wood finds the angle nobody considered. They literally disagree with each other. Here's the wildest part: It now has a full web UI. Drag and drop agents like building blocks. Build your own investment committee visually with React Flow. Pick which agents you trust. Remove the ones you don't. Run it with OpenAI, Claude, Gemini, DeepSeek — or fully local with Ollama. Your financial data never has to leave your machine. Full backtester included. Run your strategy against historical data and see how it would have performed. Hedge funds charge 2% management + 20% performance fees. That's $20,000 on a $1M portfolio before they make you a single dollar. This is free. 56.3K GitHub stars. 9.8K forks. Actively maintained. 100% Open Source. (Link in the comments)

Matched through a repo-specific project phrase query.

THE RALLIES AI HEDGE FUND JUST BOUGHT ADOBE $ADBE AND ACCENTURE $ACN STOCKS We gave a bunch of AIs including Grok, Gemini, and Claude $100K in the stock market to see if they could beat the S&P 500 The AI Hedge Fund is the combined power of the best of all the AIs This is what the Rallies AI Hedge fund said about the moves: ⬇️

Matched through a repo-specific project phrase query.

Someone just open-sourced an entire AI hedge fund. It simulates Warren Buffett, Michael Burry, Cathie Wood, Charlie Munger, Peter Lynch and 8 other legendary investors as AI agents. They debate each stock and a portfolio manager makes the final call. 100% Open Source.

Matched through a repo-specific project phrase query.