鸟哥 Bluebird Labs

Buffett's 70 years of investment wisdom has been transformed into a knowledge map by one person in two days. It's recommended that every investor should take a look at it!

This website has broken down all 98 shareholder letters from Buffett from 1956 to 2025 into an interactive knowledge base👇

📊 Data Volume:

- 98 letters (partner letters + Berkshire shareholder letters, spanning 70 years)
- 49 investment concepts (intrinsic value, moat, capital allocation…)
- 61 companies (from Coca-Cola to Apple)
- 7 key figures (Munger, Graham…)
- 4726+ cross-references

The strongest aspect is: each concept, each company, and each individual is linked back to the original source, allowing for one-click tracing. This is not a second-hand summary; it directly points to Buffett's original words.

For example, the concept of "intrinsic value" appears 83 times in the 98 letters, and the context of each citation can be explored.

There's also a core idea map that visualizes the connections between Buffett's investment philosophies.

For those investing, researching, or wanting to systematically learn Buffett's thoughts, this is ten times more efficient than reading the original texts.

🔗 https://buffett-letters-eir.pages.dev

10 websites that cryptocurrency enthusiasts must bookmark; missing one could lead to losing money!

1️⃣ Macroeconomic Data: https://en.macromicro.me
CPI, interest rates, employment… not looking at macro when trading crypto is like driving with your eyes closed.

2️⃣ BTC Halving/Shutdown Price/Fear Index: https://fuckbtc.com
The name is crude, but the data is comprehensive; a must-see for BTC cycles.

3️⃣ Financing Information Inquiry: https://rootdata.com
Which project took whose money, just check it out.

4️⃣ Contract Data/Liquidation Map: https://coinglass.com
Funding rates, long-short ratios, liquidation heatmaps; if contract players don't look at this, they're just giving away money.

5️⃣ DeFi Data: https://defillama.com
TVL, protocol rankings, yields; essential for DeFi players.

6️⃣ On-chain Data Analysis: https://dune.com
Custom dashboards; a must-have tool for on-chain detectives.

7️⃣ Token Unlocking: https://tokenomist.ai
Which coin unlocks how much at what time; know in advance to act early.

8️⃣ Smart Money Tracking: https://intel.arkm.com
Whale addresses, institutional wallet movements; follow the smart money.

9️⃣ Airdrop Projects: https://airdrops.io
Airdrop information aggregation; don't miss out on free money anymore.

🔟 Market Data Overview: https://coingecko.com
Coin information, market cap rankings, on-chain data; the most basic yet the most important.

Your experience is being "distilled" into AI by the company, did you know?

Recently, Twitter exploded—after "colleague.skill" became popular, more and more companies are requiring employees to write their work experience into AI Skills documents.

Simply put, it's about taking what’s in your head, feeding it to AI, and then you can be replaced.

It's called "knowledge sedimentation", but in reality, it's "knowledge robbery".

Some people can't sit still.

A new project just came out on GitHub—anti-distill (anti-distillation), specifically to deal with this 👇

🛡️ What it does:

1️⃣ You provide the company's Skills document, and it helps you automatically "dehydrate"—with a complete structure, professional terminology, looking somewhat legitimate, but the core experience is all replaced with "correct nonsense"
2️⃣ At the same time, it generates a private backup, real judgment, lessons learned, and networking resources, all kept for yourself
3️⃣ Three levels of intensity to choose from: mild retains 80%, moderate 60%, and heavy only retains 40%

For example:
"The TTL must be set for Redis keys, otherwise the PR will be rejected" → after distillation becomes "Caching follows team norms"

Technically completely correct, but of no value in practice. This is the essence of anti-distillation.

To be honest, this project hits the pain points of all workers—your experience is your core asset, why should you give it away for free in exchange for a layoff notice?

In the AI era, protecting your irreplaceability is more important than learning any tool.

🔗 https://github.com/leilei926524-tech/anti-distill

Watch videos to learn languages (English, Japanese), this is the way AI should be opened! It's not boring! Any video content you like!

A 3.5k⭐ open-source project on GitHub—LLPlayer, a video player designed specifically for language learning, fully featured👇

🎬 Core capabilities:

1️⃣ AI real-time subtitle generation—based on Whisper, any video/audio can automatically generate subtitles without needing a subtitle file
2️⃣ Bilingual subtitles displayed on the same screen, supporting both text and image formats
3️⃣ Real-time translation—access multiple engines like Google, DeepL, Ollama, OpenAI, and can use LLM for context-aware translation, much more accurate than regular machine translation
4️⃣ OCR recognition—hardcoded subtitles can also be converted to text in real time, powered by Tesseract
5️⃣ Subtitle sidebar supports search, jump, and word lookup, learn while watching
6️⃣ Built-in yt-dlp, direct playback and learning from YouTube

💡 The most critical point: ASR and OCR run entirely locally, ensuring privacy and security, with support for NVIDIA GPU acceleration.

To be honest, using it to binge-watch American dramas to learn English or Japanese variety shows to learn Japanese is more practical than any language learning app. It's free and open source, not charging you a penny.

The only regret—currently, it only supports Windows.

🔗 https://github.com/umlx5h/LLPlayer

In Q1 2026, who is truly the king of the crypto world? The latest report!

CoinGlass's latest report provides the answer—Binance, crushing all dimensions.
The number one exchange in the universe is not just given away!

This is not bragging; the data speaks👇

1️⃣ Derivatives market share 34.9%, leading by a wide margin

2️⃣ Contract open interest (OI) accounts for 29.9%, with the deepest capital accumulation

3️⃣ User asset share 73.5%—note that over 70% of the money is in Binance

4️⃣ Liquidity depth far ahead, with the lowest slippage for large trades

What does this mean?

The total global crypto trading volume in Q1 is $20.57 trillion, with Binance alone taking one-third. User assets are even more exaggerated; all other exchanges combined do not even reach a fraction of it.

In short, the ultimate competition in the crypto industry has long been settled. What remains is just the question of who will be second.

The only variable? On-chain. DEXs like Hyperliquid are quietly encroaching on CEX territory, but it is still early to shake Binance's throne.

🔗
https://coinglass.com/en/learn/2026-q1-mktshare-report-en

Today, someone still privately messaged me asking how to download videos from X. This open-source tool can download videos from all over the internet!

No ads, no tracking, no paywalls—this is how a download tool should be.

Cobalt, a pure media downloader—paste the link, get the file, and leave. 39.4k Star.

Supported platforms👇

🎬 YouTube
📸 Instagram
🎵 TikTok
🐦 Twitter/X
📱 Reddit
🎧 SoundCloud
📹 Vimeo
🎶 Spotify
📎 VK
…and more

It doesn't try to do everything, it focuses on one thing: you give it a link, it gives you the original file.

Differences from other download tools:

1⃣ Zero ads, zero tracking, zero paywalls—doesn't profit off you
2⃣ Works as a proxy, doesn't cache your content
3⃣ Only processes publicly accessible content, just like what browser developer tools can do
4⃣ AGPL-3.0 open source, fully transparent code

Written in Svelte + JavaScript + TypeScript, front-end and back-end separated, supports self-hosting. 3,248 commits, very actively maintained.

For content migration, material collection, and data backup, this is much cleaner than those download sites filled with pop-ups.

⭐ 39,400 | 🍴 3,300

🔗 github.com/imputnet/cobalt

After crawling 22,000 articles from the Wall Street Journal, I wanted to use news sentiment to predict the stock market, but the result was quite a slap in the face.

A data science project by Philippe Heitzmann: using Python + Selenium to crawl 22,772 full texts of WSJ from 2019 to 2020, performing NLP sentiment analysis to see if it can predict the S&P 500 trend.

Two questions were studied👇

📰 Article sentiment vs reader engagement
→ Articles with stronger negative sentiment have more comments (VADER negative scores are significantly at 1% level)
→ But the overall predictive power is very weak, with an R² of only 0.014

📉 Article sentiment vs market trend
→ Tested four time windows from the day to T+3
→ Result: R² is approximately 0.01, with almost no predictive capability
→ The conclusion is quite honest — relying solely on WSJ sentiment to predict the stock market is basically useless

Technology stack: Selenium web scraping, VADER + TextBlob sentiment analysis, Jupyter Notebook, R Shiny visualization, Docker deployment.

What is worth noting about this project is not the result, but the process — it falsified a hypothesis that many people firmly believe: "news sentiment can predict stock prices."

Those who engage in quantitative analysis know that using sentiment factors alone is essentially useless; it has to be combined with other factors to be potentially effective. This project serves as the best counterexample.

⭐ 56 | 🍴 15

🔗 github.com/philippe-heitzmann/WSJ_WebScraping_NLP

The quantitative investment platform personally developed by Microsoft, used by quant teams on Wall Street!

Qlib covers the entire quantitative research chain from factor mining to model training to backtesting to live deployment.

It's not a toy-level demo; it's a production-level framework produced by Microsoft Research.

Core capabilities👇

🧠 Built-in a bunch of SOTA models — LightGBM, Transformer, TabNet, HIST, IGMTF, KRNN, ready to use out of the box

📈 Full process coverage — Data processing → Model training → Backtesting → Portfolio optimization → Online deployment

🔄 Market drift adaptation — Concept Drift technology to cope with dynamic changes in the financial market

⚡ High-frequency trading support — 1-minute K-line data, HFT strategies can run

🤖 RD-Agent — The latest integrated LLM automation system, AI automatically mines factors and optimizes models

`pip install pyqlib`, Python 3.8-3.12, Docker images are also available.

Suitable for those who truly engage in quantitative analysis, not for onlookers.

⭐ 40,100 | 🍴 6,300

🔗 github.com/microsoft/qlib

FinGPT, another major project of the AI4Finance Foundation—an open-source financial large language model, primarily addressing one issue: enabling ordinary people to access Wall Street-level financial AI.

The difference from BloombergGPT: Bloomberg spent 3 million dollars training a financial LLM from scratch, whereas FinGPT's approach involves lightweight fine-tuning (LoRA) on an open-source large model, with each iteration costing less than 300 dollars, and it can update the latest financial data weekly/monthly.

19k Star, included in NeurIPS 2023, not an amateur project.

Five-layer architecture👇

1⃣ **Data Source Layer** — Real-time market information retrieval
2⃣ **Data Engineering Layer** — NLP processing of high-frequency financial data
3⃣ **Large Model Layer** — Fine-tuning for timeliness
4⃣ **Task Layer** — Benchmark testing and evaluation
5⃣ **Application Layer** — Production environment deployment

⭐ 19,000+ | 🍴 2,700+

🔗 github.com/AI4Finance-Foundation/FinGPT

AI helps you write investment research reports, with full analysis of 15 types of charts and three financial statements, quality comparable to broker research reports.

FinRobot, the financial AI Agent platform produced by AI4Finance Foundation, is dedicated to the task of investment research. It is not just a shell of ChatGPT that lets you ask questions; it is a system that can automatically generate complete investment research reports.

The biggest difference from other "AI stock trading" projects: it has academic paper support (arXiv: 2405.14767), and the architectural design is serious.

Core capabilities👇

📊 **FinRobot Pro (Web version research report assistant)**
→ Automatically generate professional-level equity research reports
→ Full analysis of three financial statements: income statement, balance sheet, cash flow statement
→ Valuation analysis: PE, EV/EBITDA, industry comparison
→ Comprehensive risk assessment
→ Multi-page HTML/PDF reports, 15+ types of charts
→ Example reports available for NVDA, MSFT, TSLA, META, etc.

Who is it suitable for:
→ Individual investors who want broker-level research reports but cannot afford analysts
→ Quantitative teams looking to use AI to speed up the investment research process
→ Academic researchers studying financial AI Agents

⭐ 6,500 | 🍴 1,100

🔗 github.com/ai4finance-foundation/finrobot

Let AI team up to work, the one-person company OPC has been open-sourced!\n\nCrewAI, a multi-agent collaboration framework written in Python, core idea: assign roles and tasks to AI, allowing them to collaborate like a real team to complete complex work.\n\nOver 100,000 certified developers are using it, it's not a toy.\n\nThe biggest difference between it and other agent frameworks👇\n\nIt does not rely on LangChain. An independent framework, lighter, faster, and with lower resource consumption.\n\nTwo working modes:\n\n🤖 Crews (autonomous teams)\n→ Each agent has independent roles, goals, and expertise\n→ Agents can make decisions autonomously and dynamically delegate tasks\n→ Suitable for scenarios that require flexible judgment—market research, content creation, data analysis\n\n⚡ Flows (event-driven workflows)\n→ Fine control over each step of the process\n→ Safe state management, conditional branching\n→ Suitable for production environments—approval flows, data processing pipelines, standardized business processes\n\nThe two can be used in combination: Crews handle parts that require creativity, Flows handle parts that require certainty.\n\nWhat can it actually do:\n\n1⃣ Market research — one agent searches for information, one agent analyzes data, one agent writes reports\n2⃣ Stock analysis — fundamental agent + technical agent + risk assessment agent collaborates to produce investment research reports\n3⃣ Content production — planning agent selects topics, writing agent drafts, editing agent reviews\n4⃣ Travel planning — destination agent checks guides, budget agent calculates costs, itinerary agent schedules\n5⃣ Recruitment process — screening agent reviews resumes, assessment agent scores, summary agent produces candidate list\n\nInstall it in one line:\n`uv pip install crewai`\n\nIn simple terms, CrewAI addresses the issue of "one AI is not enough, we need a group of AIs to work together." A single agent is suitable for simple tasks, but truly complex work requires division of labor and collaboration—this is the battlefield of CrewAI.\n\n⭐ 48,000+ | 🍴 6,500+\n\n🔗 github.com/crewAIInc/crewAI

Let Buffett, Munger, and Cathie Wood help you analyze a stock at the same time; this open-source project can do it!

ai-hedge-fund, an open-source project that simulates a hedge fund research team using 18 AI Agents, 50k Star.

Its idea is very bold: instead of creating an analysis model, it creates an entire investment team—each Agent plays a real investment master, using their investment philosophy to analyze the same stock.

12 investment master Agents👇

🧠 Warren Buffett — Value investing, moat thinking
🧮 Charlie Munger — Multidisciplinary thinking model
📊 Ben Graham — Margin of safety, Graham formula
🔥 Cathie Wood — Disruptive innovation, growth stocks
🏦 Bill Ackman — Activist investing, event-driven
💀 Michael Burry — Contrarian investing, deep value
📈 Peter Lynch — Finding ten-baggers in life
🔍 Phil Fisher — Chitchat method research
💎 Mohnish Pabrai — Low risk, high return
🐘 Rakesh Jhunjhunwala — Indian stock god, emerging market perspective
💰 Stanley Druckenmiller — Macro hedge
📉 Aswath Damodaran — Valuation guru

Plus 6 professional Agents:

4⃣ Valuation analyst, sentiment analyst, fundamental analyst, technical analyst
2⃣ Risk manager, portfolio manager

The workflow is as follows: 12 masters independently analyze → 4 professional Agents supplement data dimensions → Risk manager reviews → Portfolio manager integrates all opinions to make the final decision.

Three ways to use it:

1⃣ Command line — Run directly with Python + Poetry
2⃣ Web interface — Full-stack UI, visualize results
3⃣ Backtester — Use historical data to validate strategy performance

Supports OpenAI, Anthropic, Groq, DeepSeek, and local Ollama multi-models. Free data covers AAPL, GOOGL, MSFT, NVDA, TSLA five.

⚠️ Must be clear: This is an educational project, does not execute real trades, does not constitute investment advice. 50k Star does not mean it can help you make money—it represents that this Agent architecture design is indeed interesting.

Use it to learn multi-Agent collaboration, learn investment analysis frameworks, definitely worth going through the code. Using it for real trading? That’s another matter.

⭐ 50,000 | 🍴 8,700

🔗 github.com/virattt/ai-hedge-fund

Wall Street is shocked!
Someone has created a Chinese version of Wall Street's multi-Agent quantitative trading system.

TradingAgents-CN, based on LangGraph, is a multi-AI Agent collaborative stock trading framework. The original version was an academic project aimed at the U.S. stock market, and this fork has made a complete adaptation for the Chinese market.

The core architecture simulates an investment bank research team:
→ Fundamental Analyst: Financial reports, valuations, industry comparisons
→ Technical Analyst: Candlestick patterns, indicator signals
→ News Analyst: Public opinion, policies, event-driven
→ Risk Manager: Position management, stop-loss and take-profit
→ Trader: Integrates all analysts' opinions for final decision-making

What has been localized:
→ Access to data from the A-share, Hong Kong, and U.S. stock markets
→ Data sources use akshare + Alpha Vantage + Finnhub + yfinance
→ Three-layer cache (Redis + MongoDB + File), automatic degradation
→ Streamlit Web interface, supports real-time progress display
→ Docker multi-architecture deployment (amd64 + arm64)
→ Complete Chinese documentation and configuration

1195 commits, maintained for nearly a year, not a small amount of code. From the commit history, it is evident that the author is seriously addressing engineering issues → event loop conflicts, import path fixes, log system reconstruction, not just a shell demo project.

But it is important to be clear: AI Agent stock trading currently has no publicly available, long-cycle verified positive return cases. The decision-making quality of the multi-Agent architecture highly depends on the reasoning capability of the underlying LLM, and the LLM's understanding of financial markets is fundamentally based on historical texts, not on market microstructure.

Suitable for quantitative research and learning about Agent architectures; do not use it directly for live trading.

⭐
22,400 |🍴4,600

🔗https://github.com/hsliuping/TradingAgents-CN

Drift Protocol was just hacked for 270 million USD, hackers certainly make money quickly!

I also want to learn hacking techniques, sharing an open-source hacking penetration toolkit:

hackingtool packages 185 security tools into a Python menu, information gathering, vulnerability exploitation, phishing attacks, post-exploitation, payload generation, with 20 categories for one-click installation. v2.0 even supports natural language search → Input "I want to scan a network", it helps you choose the tools.

This repository has 57,000 stars, originally intended for security researchers. But the Drift incident highlights a harsh reality: the attackers' toolchain has become highly engineered, while the defense's response speed lags far behind. Hackers were lying in wait on the chain three weeks ago, and only detonated today.

Open-source security tools are a double-edged sword → White hats use them for auditing and fortification, black hats use them for reconnaissance and breakthroughs. The difference lies in who acts first.

hackingtool is suitable for CTF players and penetration testers with legal authorization as a toolbox, and is not suitable for bad deeds.

⭐57,149 |
🍴6,295

🔗
https://github.com/Z4nzu/hackingtool

Want to do quantitative portfolio optimization? This Python library has open-sourced Wall Street-level tools.

Riskfolio-Lib, a professional portfolio optimization and quantitative asset allocation library. The covered topics are very hardcore: efficient frontier, risk parity, CVaR optimization, maximum drawdown model, principal component regression, duration matching, Sharpe ratio optimization... basically, it implements all the portfolio optimization methods that can be used in quantitative investing.

It is based on CVXPY for convex optimization solving, not a toy-level backtesting framework, but a real tool for asset allocation decision-making. Suitable for quantitative researchers, asset management practitioners, or anyone who wants to systematically learn modern portfolio theory.

3800+ stars, 600+ forks, indicating a real user base in the quantitative circle.

🔗https://github.com/dcajasn/Riskfolio-Lib

Let your AI Agent instantly transform into a Wall Street analyst, this collection of financial skills is worth keeping.
With this news coverage, it's enough; OpenClaw can handle it!

Awesome Finance Skills, a plug-and-play financial analysis agent skill package covering 8 major modules:

→ Real-time news aggregation: Financial Associated Press, Wall Street Insights, Weibo, Polymarket, and more than 10 news sources.

→ A-shares/Hong Kong stocks data: stock code search, OHLCV historical data.

→ Sentiment analysis: market sentiment scoring based on FinBERT/LLM (-1.0 ~ +1.0).

→ Time series forecasting: Kronos model predictions, automatically adjusting based on news sentiment.

→ Logic link visualization: automatically generating market impact transmission chain diagrams.

→ Investment signal tracking: evolution tracking of signal reinforcement/weakening/refutation.

→ Research report generation: planning → writing → editing → charts, one-stop report.

→ Search and RAG: Jina/DDG/Baidu multi-engine retrieval.

Compatible with mainstream agent frameworks like Claude Code, Codex, OpenCode, Antigravity, install with one command:
`npx skills add RKiding/Awesome-finance-skills@alphaear-news`

For those doing quantitative research or financial content, this package saves a lot of work connecting to data sources. There is a free online demo to try first.

⭐645 |
🍴84 | Python

🔗https://github.com/RKiding/Awesome-finance-skills

The new feature released by Binance Web3 wallet is amazing. Meme culture is all about celebrity narratives, and the tags of these well-known figures are crucial. For beginners, it’s easy to see which Meme relates to which celebrity, saving a lot of time when looking for angles in the future!\n\nThe Binance Web3 wallet web version has added a "topic tag" feature to its Hot Radar. It focuses on high-impact Twitter accounts and automatically captures market trends, with 5 selected tag categories to filter out noise with one click. In simple terms, it helps you quickly locate what narratives are currently trending amidst the information flood.\n\nThis feature definitely deserves a thumbs up for the product manager of the Binance wallet. A few days ago, I shared an article titled "Afeng revealed the secrets of making tens of millions a year with Memes, everyone should read it carefully a few times. You might not make that much, but at least you won't lose money!" which illustrates this point!\n\nEntry:\nOpen Binance wallet: https://web3.binance.com/referral?ref=BNBSOL\nTrench---Hot Radar---Filter

Traditional education is going to be revolutionized by AI! In the future, it might really not be necessary to pay tuition fees!

A team from Tsinghua University has developed an open-source Agent that supports OpenClaw lobsters, which can convert any knowledge and learning materials into AI interactive tutorial classes. AI becomes your teacher, and there are AI classmates interacting with you, plus you can export course materials (which teachers can use for lectures)!

Memory improved tenfold, saving hundreds of thousands in tuition fees, and self-learning can be used in the future. It makes it easier for traditional school teachers to create course materials!

Warehouse address:
https://github.com/THU-MAIC/OpenMAIC?tab=readme-ov-file

Two mini versions of OpenClaw lobster, optimizing the Opencalw code to the extreme, can run on low-spec idle devices, and someone has already deployed OpenClaw on Raspberry Pi, a favorite among geek players!
 
ZeroClaw
http://github.com/theonlyhennygod/zeroclaw
Rust re-written OpenClaw, memory reduced by more than 190 times
• Size: 28MB → 3.4MB (8 times)
• Start: 5.98s → 0s
• Memory: 1.52GB → 7.8MB (194 times difference)
 
PicoClaw
http://github.com/sipeed/picoclaw
Running core functions on a 10MB RAM RISC-V with 1% of OpenClaw's code and 1% of its memory. Lobsters can be deployed on a $10 development board and can run Linux systems.

A skill information intelligence station that can filter important content from thousands of information source RSS feeds, including X/Twitter, Reddit, etc., pushing to you every 4 hours and daily, supporting Chinese, open source and free!
 
Repository address: https://github.com/kevinho/clawfeed
OpenClaw lobster installation command: clawhub install clawfeed