ai-berkshire: a multi-agent value investing framework that simulates legendary investors to produce decision-grade research reports

What it solves

AI Berkshire addresses the lack of decision-grade quality and discipline in standard AI-generated investment analysis. While general LLMs often provide balanced but vague "on the one hand... on the other hand" responses, this framework forces AI to provide concrete conclusions, specific price ranges, and rigorous financial validation to enable actual investment decision-making.

How it works

The project provides a collection of structured "Skills" (commands) compatible with Claude Code and Codex. It transforms a single user into a virtual investment team by employing a multi-agent architecture:

• Multi-Perspective Agents: It simulates a team of four value investing masters (Warren Buffett, Charlie Munger, Duan Yongping, and Li Lu), each analyzing a company from a distinct angle (e.g., business essence, moat, inverse thinking, and long-term certainty). These agents are designed to challenge each other to eliminate blind spots.
• Structured Workflows: It uses predefined checklists and funnels (e.g., the "Mirror Test") to ensure consistent depth and format across different companies.
• Financial Rigor Tools: To prevent LLM calculation errors, it uses a Python-based tool (financial_rigor.py) utilizing decimal.Decimal for precise calculations and cross-referencing data from multiple independent sources.
• Layered Design: It separates functionality into a Skill layer (19 specific entry points), an Agent layer (parallel scheduling of masters), and a Tool layer (precise calculation and retrieval).

Who it’s for

It is designed for investors who want professional-grade, structured value investment research without needing a full human research team, and for those using Claude Code or Codex to automate their financial analysis.

Highlights

• Decision-Oriented Output: Forces a "Pass/Fail/Grey" conclusion rather than a generic summary.
• Multi-Agent Conflict: Uses four distinct investment philosophies to create tension and avoid confirmation bias.
• Anti-Bias Mechanisms: Includes information richness ratings (A/B/C), Munger-style inverse testing, and a "fast-reject" red-line list.
• Comprehensive Skill Set: Offers 19 specialized tools covering deep research, earnings reviews, industry screening, portfolio management, and news attribution.
• Financial Precision: Implements strict verification for market caps and valuations to avoid common LLM arithmetic hallucinations.