AI Berkshire — Value Investing Research Framework for AI Agents

Introduction

AI Berkshire is a multi-agent investment research framework that applies value investing methodology through LLM-powered analysis. It models distinct investor perspectives—Buffett, Munger, Duan Yongping, and Li Lu—as separate agents that independently evaluate a stock, then subjects their conclusions to adversarial debate to surface blind spots.

What AI Berkshire Does

• Runs multiple analyst agents, each embodying a different value investing philosophy
• Fetches fundamental data including financial statements, valuation ratios, and industry metrics
• Produces structured research reports with bull and bear cases for a given stock
• Uses adversarial analysis where agents challenge each other's assumptions and conclusions
• Generates a synthesis report aggregating findings and highlighting agreement and disagreement

Architecture Overview

The framework follows a multi-agent pipeline. A data layer fetches financials from public APIs. Each analyst agent applies its own lens—margin of safety, competitive moats, capital allocation, or growth-at-reasonable-price. After independent analysis, a debate module orchestrates adversarial exchange. A synthesis agent produces the final report, weighting conclusions by evidence strength.

Self-Hosting & Configuration

• Requires Python 3.10+ and an LLM API key (supports OpenAI, Anthropic, and local models)
• Financial data sources are configurable; defaults to free APIs for US equities
• Analyst agent personas and their evaluation criteria are defined in YAML config files
• Debate rounds, temperature settings, and output format are adjustable via CLI flags
• Reports can be output as Markdown, JSON, or HTML

Key Features

• Multi-perspective analysis that avoids single-viewpoint bias in research
• Adversarial debate mechanism that stress-tests investment theses before conclusion
• Configurable analyst personas so users can add or modify investment frameworks
• Structured output with clear bull/bear cases and confidence levels
• Support for both cloud and local LLMs to keep research workflows private

Comparison with Similar Tools

• FinRobot — general financial AI agent; AI Berkshire focuses specifically on value investing with adversarial multi-agent debate
• OpenBB — financial data terminal; AI Berkshire adds LLM-powered analytical reasoning on top of raw data
• GPT Researcher — general research agent; AI Berkshire is domain-specific with investor persona modeling
• Stock Analysis AI agents — typically single-agent; AI Berkshire uses multi-agent adversarial review

FAQ

Q: Does it provide buy or sell recommendations? A: It produces research reports with bull and bear cases. It is a research tool, not financial advice.

Q: Which financial data sources does it use? A: It defaults to free public APIs for US equities. Users can configure additional data sources in the settings.

Q: Can I add my own analyst persona? A: Yes. Analyst agents are defined in YAML files. You can add new personas with custom evaluation criteria.

Q: Does it support non-US stocks? A: Yes. Configure the appropriate data source adapters for other markets.

Sources

• https://github.com/xbtlin/ai-berkshire