AI Berkshire
A set of AI prompt-based value investing research skills for Claude Code and Codex.
- Stars
- 17k
- Forks
- 2.5k
- Licence
- MIT
- Language
- HTML
- Last update
- 4 days ago
- Type
- AI model / agent
What it is
AI Berkshire is a collection of investment research "skills" (structured prompts and workflows) that run inside Claude Code or Codex. It is built around the investing methods of Buffett, Munger, Duan Yongping and Li Lu, and includes multi-agent workflows where several agents research a company in parallel and then a lead agent combines their conclusions. The README is largely in Chinese and focuses on Chinese and US stocks.
- Run a deep research report on one listed company using the investment-research skill
- Quickly screen a list of companies with the quality-screen skill before deeper work
- Review an earnings report with the earnings-review skill
- Track whether an investment thesis still holds with the thesis-tracker skill
Great for
- Chinese-speaking investors who want a structured value investing checklist applied to specific companies
- People already using Claude Code or Codex who want repeatable research templates
- Analysts who want to compare several companies against the same criteria
- Readers who want to see how an AI-driven research process is organized
Not ideal for
- Anyone wanting guaranteed or verified investment returns: the README's performance figures are self-reported and it warns that past returns do not predict future ones
- Beginners without an AI coding client subscription, since the skills depend on Claude Code or Codex
- People who need a licensed financial advisor's personal recommendation; the project is a research aid, not advice
How to get it
- 1
Install with a package manager
You'll need: Node.js (required for the npm install of Claude Code; version not stated in README) · Claude Code or Codex account
Install Claude Code
npm install -g @anthropic-ai/claude-codeInstall Codex via npm
npm install -g @openai/codexVerify Codex installation
codex --version - 2
Run it
- Install Claude Code or Codex using the commands in the README
- Clone the repository: git clone https://github.com/xbtlin/ai-berkshire
- Open the repository folder in your AI client
- Run a skill such as /investment-research with a company name or stock code
- 3
Get the most out of it
- Start with /quality-screen or /investment-checklist to filter companies before running slower deep research
- Use /financial-data to cross-check key figures against two independent sources, as the README recommends
- Use /thesis-tracker after buying to keep checking whether your original reasoning still holds
- Browse the published reports on the reading site and the reports index to see example outputs
Downloads
Strengths
- Gives the same research structure every time, so outputs are easier to compare across companies
- Organizes 22 named skills by task, such as deep research, earnings review, and portfolio management
- Open source under MIT with a large public following and a published report archive
Trade-offs
- Output quality depends on the underlying model and on data sources the README does not fully describe; the README itself notes higher-stakes analysis benefits from stronger models
- The track-record claims are shown as screenshots and a table, and cannot be independently verified from the repository alone
- Documentation is mainly in Chinese, with the main market focus on Chinese and US equities
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| ai-berkshire (this) | 17k | AI model / agent | intermediate | 4 days ago | Chinese-speaking investors who want a structured value investing checklist applied to specific companies |
| hermes-agent | 260k | AI model / agent | intermediate | 4 days ago | Developers who want a personal agent they can host on their own server or machine |
| frontend-slides | 31.7k | AI model / agent | beginner | 5 months ago | Non-designers who want good-looking web presentations without writing CSS or JavaScript |
| ECC | 295k | Dev tool | intermediate | 4 days ago | Developers who already use Claude Code and want a shared workflow for planning, testing and review |
| OmniRoute | 83.8k | App | intermediate | 4 days ago | Developers who use several AI coding tools and want one place to configure providers |
| learn-claude-code | 79.7k | Learning resource | intermediate | 10 days ago | Developers who want to understand how coding agents like Claude Code are structured |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.