🔥 ✨Vibe-Trading✨ just crossed 2K stars!
The AI trading agent that actually gets smarter every session — natural language in, executable strategies out, across global markets.
🌟 github.com/HKUDS/Vibe-Tra…
What we just shipped ↓
69 bundled skills + unlimited user-created × 29 agent swarm presets × 8 backtest engines.
100% MIT.
🌟 github.com/HKUDS/Vibe-Tra…
Huge thanks to the open-source community for all the support 💪
🤖 13 LLM providers — OpenAI, DeepSeek, Gemini, Z.ai, MiniMax, Ollama (local) & more
🔗 MCP plugin for Claude Desktop, Cursor & OpenClaw
🌐 5 data sources with zero-config auto-fallback — no API keys needed to start
📦 pip install, Docker, or ClawHub
this shouldn't be free.
Vibe-Trading is an open-source AI trading agent that ships with 64 finance skills, 29 specialist swarm team presets, cross-market backtesting, and a full quant analysis toolkit.
The architecture is what makes it different from everything else in this space.
It's not a wrapper around one model with a few finance prompts. It's a DAG-based multi-agent system where specialized agents collaborate, debate, and hand off between each other while you watch the entire reasoning process stream in real time.
You get:
> Technical analysis across Ichimoku, harmonic patterns, Elliott Wave, SMC, and 60+ other setups
> Quant tools: factor IC/IR analysis, quantile backtesting, Black-Scholes, full Greeks, portfolio optimization via MVO, Risk Parity, and Black-Litterman
> Alt data: social sentiment, behavioral finance signals, macro regime detection, sector rotation
> Crypto desk: perp funding basis, liquidation heatmaps, stablecoin flows, DeFi yield, token unlock tracking
> Full CLI with a TUI, a FastAPI web server, and a React frontend
HK/US equities and crypto data are completely free. Docker deploy takes 2 minutes.
github.com/HKUDS/Vibe-Tra…
MIT License. 100% Opensource.
Introdcing Vibe-Trading: Your Personal Trading Agent
💡 Vibe-Trading integrates AI-powered research automation with multi-agent swarm intelligence, to transform natural language trading ideas into executable strategies across global markets.
It offers comprehensive backtesting, quantitative analysis tools, and real-time portfolio optimization—all accessible through simple conversational without requiring coding.
Key Capabilities:
• Strategy Generation — Automatically writes trading code from your ideas
• Smart Data Access — Pulls market data from multiple sources seamlessly
• Performance Testing — Tests your strategies against historical market data
• Expert Teams — Deploys specialized AI agents for complex research tasks
• Live Updates — Watch the entire analysis process in real-time
GitHub: github.com/HKUDS/Vibe-Tra…#VibeTrading#HKUDS
上篇帖子之后,看到还是少数不怀好意的人用「虚假统计学」来诋毁😅。本来以为这么明显的统计学谬误,谣言会不攻自破,因为大家都是很优秀的researcher和审稿人。但感觉还在持续发酵,我觉得还是有必要写一篇《通俗统计学》的科普小帖子,来详细解读一下这些metrics和values,以及原po所得出的「结论」有多离谱...
原帖最重要的部分就是这张「时序分析」,一共也就两份数据+一份解读,我们一项一项来看:
A. 「前1500个样本的趋势分析」:原po选取了两个项目来做这个分析,这个分析本来就很奇怪:
a. 为啥是前1500?前1500又会怎么样?
b. 为啥是这两个项目?其它项目呢?
c. 看了一眼图表我终于懂了,原来这两个项目都符合「前X天的star数恰好等于1500这个数字」。第一个项目X=3,第二个项目X=2。那作者想用这个表达什么呢?看完结论我都震惊了:
「原作想表达,过了X天之后,这两个项目第X+1天的star都“归零”了」
哇去!你只取前1500个star,然后前X天恰好等于1500,然后你说「第X+1天“没有star”了,因为1500个star被“用完了”」这是什么神人逻辑。这是甚至不舍得用比较强的模型来做vibe统计学吗,看起来像是7B模型都不会犯的逻辑错误。你作者甚至不愿意去star统计网站上看一下到底第X+1天的实际star是多少吗?😅
再类比一下,这就好像说,你统计一个人“《前1000块钱》的消费记录”,然后发现:
前3天刚好花了《1000块》
第4天《没有消费》
于是你得出结论:
「这个人第4天不花钱了,因为1000块已经被用完了」
总结为:尬。。
B. 「账号重叠分析」
a. 只分析4个repo,一共5542个“唯一账号样本”,意思应该是取并集之后去重
b. 标黄的:227个账号同时对「2」个repo同时star。我本人反应:soooo what... 4.1%... so what???!
b. 标橙、红的:更少了,25个账号、6个账号。我请问:所以呢???
c. 看完这个原作的结论我还是没理解:为什么不能有4.1%的人同时收藏了任意2个不重复的repo,为什么不能有千分之四点人同时收藏3个,不能有千分之一的人同时收藏4个。我感觉我都不用分析了,简直无厘头啊。。!
这就好像你说:
有4%的人同时在YouTube和TikTok上点赞过视频
于是你得出结论:
「这些平台互相刷数据」
???
正常人本来就会同时用多个平台啊。
还像你统计发现:
有4%的人同时关注了3个博主
于是你说:
「这3个博主的粉丝是买来的」
那问题来了:
难道一个人一辈子只能关注一个人???
写到这里,突然觉得这篇帖子实际想用数据帮证明我们其实就没买star吗?这...
C. 总结
这篇帖子的vide coder甚至不是Claude/GPT系列模型,目测是Qwen-2B都不如. 原po,你找数据尬黑都不愿意花钱吗。有时震惊如果大家这么明显的问题都看不出来,还怎么去写research paper、当审稿人呢?是不是这也是当前投稿审稿乱象的一个投影😵
欢迎任何关于实际数据和大众统计学的热烈讨论😁😁