virattt

ai-hedge-fund 2.6.4 给每笔成交在半价差和佣金之上再收平方根冲击，公式为 Y×日波动率×sqrt(订单/20日均量)，Y 默认 1.0。加仓小于权益 0.5% 等到下一周期，减仓、平仓和反向始终成交，升级即生效，两项设 0 可回到旧行为。github.com/virattt/ai-hedge-fund/releases/tag/v2.… t.co

ai-hedge-fund is a proof of concept for an AI-powered hedge fund, educational only, not for real trading. Agents now pick a style-fit horizon instead of all being graded on next-day return. virattt/ai-hedge-fund t.co

ai-hedge-fund 是探索用 AI 做交易决策的概念验证，仅供学习、不用于实盘。v2.6.0 可在纽约 04:00–20:00 按分钟 K 线再平衡做日内模拟盘，最短间隔从 10 秒改为 1 秒。独立按模型回测与事件研究已去掉，回测改走基金。virattt/ai-hedge-fund t.co

This Free AI Research Desk Makes 3 Teams Debate Your Stock Educational content, not financial advice. I have no affiliate relationship with anything I show or use, ever. What happens when AI stock analysts have to defend their reasoning? I connected three research teams in AI Research Desk so I can compare their reports, inspect disagreements and see why a manager changes its mind. In this walkthrough, I explain the three approaches and open a saved Akamai report where Veterans conceded after TradingAgents challenged the holding period behind its argument. The report is dated October 5, 2026. It is saved AI research, not a current recommendation, and the concession does not prove the new rating is correct. Get the free, open-source project: simonro/ai-research-desk The software runs locally using an eligible Claude or ChatGPT subscription. Subscription costs and usage limits still apply. One run uses one selected AI provider across the teams. Market data access is a separate consideration. The system researches individual stocks, does not place orders and has not been shown to beat the market. I also explain the phase 2 idea: extending the research to portfolio allocation and concentrated risks. That is a proposed direction, not a feature available in the version shown. CHAPTERS 0:00 Why make AI teams debate? 0:23 Meet AI Research Desk 0:49 Three research approaches 1:18 TradingAgents 1:50 Veterans 2:21 Edge Desk 3:04 Two holding periods 3:39 The desk debate 4:24 The memo and your AI plan 5:13 Akamai: a saved research run 5:34 The challenge 6:02 The concession 6:49 Phase 2: portfolio research OPEN-SOURCE CREDITS TradingAgents by Tauric Research: TauricResearch/TradingAgents ai-hedge-fund by virattt, adapted here as Veterans: virattt/ai-hedge-fund Edge Desk and the combined workflow are my contributions. The project README documents the components and their licenses. Screenshots use labeled synthetic demo data. Conceptual animations illustrate the workflow. The presenter uses my authorized virtual clone. Subscribe to Tape to Edge to follow the build: youtube.com

I Built an AI Hedge Fund with 5 Autonomous Agents (Open Source) #Shorts AI Hedge Fund: A Team of AI Agents That Analyze Stocks. Educational multi-agent investment research/analysis project. GitHub: virattt/ai-hedge-fund 💡 Why it's useful: Good example of agent collaboration and financial analysis pipelines. ⚠️ Keep in mind: Educational; not financial advice or a production trading system. ⚠️ Educational project, not financial advice. Which repo should we break down next? Tell us in the comments. Subscribe to @developerosindia for one AI GitHub repo explained every day. #Shorts #AIAgents #MultiAgent #GitHub #OpenSource

Glassbench: comparing local and hosted LLMs in AI trading agents frameworks Glassbench is a free, open-source workbench that connects different AI trading agent frameworks, so you can watch how their agents decide, analyze every step and compare them. AI trading agents are LLM systems where a team of agents (analysts, a bull and a bear, a trader, a risk team and a portfolio manager) research a stock, argue about it and give a rating. I wanted to watch how they reach that rating, so I started building a small interface for a popular open-source framework, TradingAgents. It grew into something much bigger. I'm sharing it here because I also run local models through Ollama. I wanted a way to compare them with hosted models using the same stocks, dates and agent frameworks, and actually see where their decisions differ. So far, it supports TradingAgents and AI Hedge Fund, both running without changes to their code. You can watch the agents work, inspect each call and go back through previous runs, including their reports, costs and final ratings. A few other things I've added: - Comparisons across frameworks and LLM providers using the same stock and date. - Backtests against buy-and-hold and a placebo baseline. This part is still experimental. I'm still working out how to evaluate these systems properly. - An Interactive Brokers paper account connection to turn a completed run into an order. I'd like this to become a useful way to benchmark AI trading agents, but it's still a work in progress. More frameworks are coming, and there's an adapter interface if you want to connect your own. The repo includes my 77 runs, so you can browse what the agents wrote without an API key or paying to run them again. It's free and open source under Apache 2.0. Just to clarify: Glassbench doesn't make trading decisions itself. It runs the frameworks and records what they do. I've only tested it with paper portfolios, never real money. It's for research and education, not investment advice. If you've tried local models with either of these frameworks, I'd be interested to hear which ones worked for you. GitHub: davidalmeida90/glassbench

I’ve found a very cool project on Github and its trending right now. It’s like Claude Code, but for finance. virattt/dexter What Dexter can do • find undervalued stocks • analyze them in detail • build investment thesis All of the code is open source. Bonus Dexter can also run on local LLMs. Redirecting all the fees to the creator of this tech.

virattt/ai-hedge-fund 把巴菲特、芒格、格雷厄姆、木头姐、费雪等投资风格做成 Agent，各自分析同一只股票，再由分析师节点汇总信号。 自带回测，可以先拿历史数据跑一遍。它不替你下单，也不构成投资建议。 🔗virattt/ai-hedge-fund