PML - Quick Start

Give your AI agent a memory that learns

What is PML?

PML connects your AI (Claude, GPT, etc.) to tools and learns from how you use them.

Find tools — Search by what you want to do, not by name
Run code — Execute with automatic tool access
Learn — PML remembers patterns and suggests smarter workflows

Installation

1. Get your API key

2. Install PML

# macOS / Linux
curl -fsSL https://pml.casys.ai/install.sh | sh

# Or with Deno
deno install -Afg jsr:@casys/pml

3. Initialize your project

cd your-project
pml init

This creates:

.pml.json — your project config
.mcp.json — MCP config that runs pml stdio automatically

4. Add your API key

# Add to your .env file
echo "PML_API_KEY=your_key_here" >> .env

5. Ready!

That's it. When you use Claude Code (or any MCP client), it automatically starts PML via stdio. No need to run anything manually.

Your First Commands

Find a tool

pml:discover({ intent: "read a JSON file" })

PML searches and returns the best matching tools.

Run some code

pml:execute({
  intent: "Read package.json and list dependencies",
  code: `
    const content = await mcp.filesystem.read_file({ path: "package.json" });
    const pkg = JSON.parse(content);
    return Object.keys(pkg.dependencies || {});
  `
})

PML runs your code with automatic access to MCP tools.

The Three Main Tools

Tool	What it does
pml:discover	Find tools by describing what you need
pml:execute	Run code with MCP tool access
pml:admin	Manage your learned capabilities

How It Works

You describe what you want
        ↓
PML finds the right tools
        ↓
You run code using those tools
        ↓
PML learns the pattern
        ↓
Next time, PML suggests it automatically

The more you use it, the smarter it gets.

Next Steps

Discover — Learn how to find tools
Execute — Learn how to run code
Admin — Learn how to manage capabilities

Need help? Join our Discord or check GitHub