LLM Integration

Configure AI assistants like Claude, ChatGPT, and Copilot to understand and use the Wendy CLI

Using LLMs with the Wendy CLI

Large Language Models (LLMs) like Claude, ChatGPT, GitHub Copilot, and Cursor are transforming how developers interact with command-line tools. However, these AI assistants may not have been trained on Wendy CLI commands, especially newer features.

The --experimental-dump-help flag solves this by providing a machine-readable JSON dump of all Wendy CLI commands, parameters, and documentation that LLMs can parse and understand.

The Problem

When you ask an AI assistant to help you with Wendy CLI commands, it might:

Not know about Wendy at all
Hallucinate commands that don't exist
Miss important flags or options
Use outdated syntax

The Solution

Run wendy --experimental-dump-help to generate a complete, structured reference of all CLI commands:

wendy --experimental-dump-help

This outputs a JSON structure containing:

All available commands and subcommands
Command descriptions and documentation
All flags, options, and positional arguments
Parameter types and default values
Whether parameters are required or optional

Example Output

{
  "command": {
    "abstract": "Wendy CLI",
    "commandName": "wendy",
    "subcommands": [
      {
        "abstract": "Run Wendy projects.",
        "commandName": "run",
        "arguments": [
          {
            "abstract": "Attach a debugger to the container",
            "kind": "flag",
            "names": [{"kind": "long", "name": "debug"}]
          },
          {
            "abstract": "Run the container in the background",
            "kind": "flag",
            "names": [{"kind": "long", "name": "detach"}]
          }
        ]
      }
    ]
  }
}

Setting Up Your Project for LLMs

To help AI assistants understand and use Wendy effectively in your project, create an AGENTS.md or CLAUDE.md file in your project root.

Creating CLAUDE.md

Create a CLAUDE.md file that instructs Claude Code (and other AI assistants) to learn about Wendy:

CLAUDE.md

# Wendy CLI Project

This project uses the Wendy CLI to deploy applications to WendyOS devices.

## Learning About Wendy

Before helping with Wendy commands, run this to learn all available commands:

\`\`\`bash
wendy --experimental-dump-help
\`\`\`

This outputs a JSON structure with all commands, flags, and documentation.

## Common Tasks

- Run an app: `wendy run`
- Discover devices: `wendy discover`
- Check device status: `wendy device version`
- Update agent: `wendy device update`
- Configure WiFi: `wendy wifi connect`

Creating AGENTS.md

For broader AI assistant compatibility, create an AGENTS.md file:

AGENTS.md

# AI Assistant Instructions

## Wendy CLI

This project deploys to WendyOS devices using the Wendy CLI.

To understand all available Wendy commands and their options, run:

\`\`\`bash
wendy --experimental-dump-help
\`\`\`

Parse the JSON output to learn:
- All subcommands (run, discover, device, wifi, apps, etc.)
- Available flags and options for each command
- Required vs optional parameters
- Parameter descriptions and defaults

## Device Information

- Target device hostname: wendyos-<device-name>.local
- Default agent port: 50051
- Debug ports: 4242 (Swift/LLDB), 5678 (Python/debugpy)

Best Practices

1. Include Device-Specific Context

Add your device hostname and any custom configuration:

CLAUDE.md

## Device Configuration

- Device hostname: `wendyos-humble-pepper.local`
- Connection: USB or LAN
- Target platform: NVIDIA Jetson Orin Nano

2. Document Your Project Type

Help the LLM understand what kind of project this is:

CLAUDE.md

## Project Type

This is a Swift application targeting WendyOS.

Build and run with:
\`\`\`bash
wendy run
\`\`\`

Debug with:
\`\`\`bash
wendy run --debug
\`\`\`

3. Reference the Dump Help Command

Always remind the LLM it can refresh its knowledge:

CLAUDE.md

## CLI Reference

If you're unsure about a Wendy command or its options, run:

\`\`\`bash
wendy --experimental-dump-help | jq '.command.subcommands[] | select(.commandName == "COMMAND_NAME")'
\`\`\`

Replace `COMMAND_NAME` with the command you want to learn about.

Using with Different AI Tools

Claude Code

Claude Code automatically reads CLAUDE.md files. Place the file in:

Your project root for project-specific instructions
~/.claude/CLAUDE.md for global instructions

GitHub Copilot

Copilot reads context from open files. Keep your AGENTS.md open or reference it in comments.

Cursor

Cursor supports .cursorrules files. You can include Wendy instructions there:

.cursorrules

When working with Wendy CLI commands, first run:
wendy --experimental-dump-help

Use this output to understand available commands and their parameters.

ChatGPT / Other LLMs

When starting a conversation, paste the output of wendy --experimental-dump-help or instruct the LLM to request it.

Filtering the Output

The full dump can be large (~4700 lines). Use jq to filter for specific commands:

# Get all top-level subcommands
wendy --experimental-dump-help | jq '.command.subcommands[].commandName'

# Get details for the "run" command
wendy --experimental-dump-help | jq '.command.subcommands[] | select(.commandName == "run")'

# Get all flags for the "device" command
wendy --experimental-dump-help | jq '.command.subcommands[] | select(.commandName == "device") | .subcommands'

Example: Complete CLAUDE.md

Here's a complete example for a WendyOS project:

CLAUDE.md

# WendyOS Voice Assistant Project

## About This Project

A voice assistant application running on NVIDIA Jetson Orin Nano using WendyOS.

## Wendy CLI

To learn all available Wendy commands, run:

\`\`\`bash
wendy --experimental-dump-help
\`\`\`

### Quick Reference

| Task | Command |
|------|---------|
| Run app | `wendy run` |
| Run with debug | `wendy run --debug` |
| Run detached | `wendy run --detach` |
| Discover devices | `wendy discover` |
| List apps | `wendy device apps list` |
| Stop app | `wendy device apps stop <name>` |
| Update agent | `wendy device update` |
| Connect WiFi | `wendy wifi connect` |

## Device Configuration

- Hostname: `wendyos-humble-pepper.local`
- Platform: NVIDIA Jetson Orin Nano
- Connection: USB-C

## Development Workflow

1. Connect device via USB
2. Run `wendy discover` to verify connection
3. Run `wendy run` to build and deploy
4. View logs in terminal
5. Use `wendy run --debug` for breakpoint debugging

Next Steps

Discover your devices
Set up the VSCode extension for integrated debugging
Manage apps on your device

LLM Integration

On this page