# Does AI Berkshire Ship Pre-Trained Models? A Complete Technical Analysis

> Discover if AI Berkshire ships pre-trained models. Learn how this technical analysis reveals it interfaces with external LLMs, not providing its own model checkpoints.

- Repository: [Xbt Lin/ai-berkshire](https://github.com/xbtlin/ai-berkshire)
- Tags: technical-analysis
- Published: 2026-07-09

---

**AI Berkshire does not distribute any pre-trained model checkpoints or binaries; instead, it provides investment-research skills that interface with external LLM providers like Claude and OpenAI.**

AI Berkshire is a research framework that packages investment-research "skills" with data-analysis tools and orchestration scripts. Unlike traditional machine learning repositories, it contains no pre-trained model weights or neural network binaries. According to the source code in `xbtlin/ai-berkshire`, the repository delivers only markdown skill definitions and command wrappers that require an existing LLM endpoint to function.

## What AI Berkshire Actually Distributes

The repository contains **investment-research skills** packaged as markdown definitions rather than neural network weights. A search for common model file extensions—including `.pth`, `.ckpt`, and `.bin`—returns zero matches throughout the codebase. Instead, the project organizes its functionality into three primary asset types:

- **Markdown skill definitions** located in `skills/*.md`
- **Generated Codex artifacts** stored under `codex-skills/*/SKILL.md`
- **Orchestration scripts** within `scripts/*` for CLI integration

These components define prompts, data-analysis workflows, and report templates, but contain no executable model parameters.

## Key Files Confirming the Architecture

Several canonical files document AI Berkshire's design as a skill-orchestration layer rather than a model provider.

### AGENTS.md

The [`AGENTS.md`](https://github.com/xbtlin/ai-berkshire/blob/main/AGENTS.md) file establishes that the **canonical source** for all functionality resides in the `skills/` directory. It explicitly defines the workflow as requiring users to run skills against an existing LLM provider—such as Claude or OpenAI—rather than loading a bundled local model. This confirms the framework's dependency on external API endpoints.

### README_EN.md

According to [`README_EN.md`](https://github.com/xbtlin/ai-berkshire/blob/main/README_EN.md), the project describes itself as a *"collection of investment research skills"*. The documentation overview never mentions distributing pre-trained model checkpoints or hosting downloadable weights, focusing entirely on skill composition and prompt engineering.

## How to Deploy AI Berkshire Without Local Models

Since no pre-trained models ship with the repository, deployment requires connecting the framework to an external LLM service. The repository provides installation scripts that create command-line wrappers for popular AI coding assistants.

### Setting Up Claude Code Integration

To execute skills using Claude, install the compatible command wrappers:

```bash

# Install the Claude-compatible commands (once)

./scripts/install-claude-commands.sh

# Execute the "investment-research" skill

claude run investment-research \
    --ticker AAPL \
    --date $(date +%Y-%m-%d)

```

The [`scripts/install-claude-commands.sh`](https://github.com/xbtlin/ai-berkshire/blob/main/scripts/install-claude-commands.sh) script defines the `claude` entry point and maps skill definitions to Claude-compatible slash commands.

### Configuring Codex Prompts

For OpenAI Codex integration, use the provided prompt installer:

```bash

# Install Codex prompts (once)

./scripts/install-codex-prompts.sh

# Invoke the generated Codex prompt

codex run investment-research \
    --ticker TSLA \
    --date $(date +%Y-%m-%d)

```

This script installs generated prompt files under `codex-prompts/` based on the markdown sources in `skills/`.

### Loading Skills Programmatically

You can also inspect skill definitions directly without running them:

```python
from pathlib import Path

skill_path = Path("skills/investment-research.md")
skill_md = skill_path.read_text()
print(skill_md[:500])   # preview the first 500 characters

```

The [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md) file contains the raw prompt logic and validation rules used by both Claude and Codex integrations.

## Summary

- **AI Berkshire ships no pre-trained models**: The repository contains zero model checkpoint files (`.pth`, `.ckpt`, `.bin`).
- **Skills are markdown-based**: All functionality resides in `skills/*.md` and `codex-skills/*/SKILL.md` files defining prompts and workflows.
- **External LLM required**: The framework requires connection to Claude, OpenAI, or compatible API endpoints to execute commands.
- **Installation scripts provided**: [`scripts/install-claude-commands.sh`](https://github.com/xbtlin/ai-berkshire/blob/main/scripts/install-claude-commands.sh) and [`scripts/install-codex-prompts.sh`](https://github.com/xbtlin/ai-berkshire/blob/main/scripts/install-codex-prompts.sh) handle CLI integration without downloading weights.

## Frequently Asked Questions

### Does AI Berkshire include downloadable pre-trained model weights?

No. AI Berkshire does not contain any downloadable pre-trained model weights or neural network binaries. The repository exclusively distributes markdown skill definitions and installation scripts that connect to external LLM providers.

### Which LLM providers are compatible with AI Berkshire?

AI Berkshire supports Claude (via Claude Code) and OpenAI's Codex/GPT models. The [`AGENTS.md`](https://github.com/xbtlin/ai-berkshire/blob/main/AGENTS.md) specification indicates the architecture is designed to work with any LLM provider that accepts the structured prompt formats defined in the `skills/` directory.

### Where are the skill definitions stored in the repository?

Canonical skill definitions reside in the `skills/` directory as markdown files (e.g., [`skills/investment-research.md`](https://github.com/xbtlin/ai-berkshire/blob/main/skills/investment-research.md)). Generated variants for Codex compatibility are stored under `codex-skills/*/SKILL.md`.

### Can I run AI Berkshire offline without an internet connection?

No. Since AI Berkshire relies entirely on external LLM APIs and contains no local model weights, an active internet connection and valid API credentials for Claude, OpenAI, or another supported provider are required to execute any skills.