baml-init

Initialize BAML Project

Set up BAML in an existing project or create a new one.

Detection

First, detect the project context:

1. Check for existing baml_src/ directory
2. Check for pyproject.toml (Python) or package.json (TypeScript)
3. Check for existing generators.baml

Actions

If No BAML Exists

For Python projects:

# Add baml-py dependency
uv add baml-py
# Or: pip install baml-py

# Create baml_src directory
mkdir -p baml_src

For TypeScript projects:

# Add BAML dependency
npm install @boundaryml/baml
# Or: pnpm add @boundaryml/baml

# Create baml_src directory
mkdir -p baml_src

Create generators.baml

Detect language and create appropriate generator:

Python:

generator target {
  output_type python/pydantic
  output_dir "../baml_client"
  version "0.76.2"
}

TypeScript:

generator target {
  output_type typescript
  output_dir "../baml_client"
  version "0.76.2"
}

Create clients.baml

// OpenAI client
client GPT4 {
  provider openai
  options {
    model "gpt-4"
    api_key env.OPENAI_API_KEY
  }
}

// Vision-capable model for images
client GPT4o {
  provider openai
  options {
    model "gpt-4o"
    api_key env.OPENAI_API_KEY
  }
}

// Default retry policy
retry_policy Default {
  max_retries 2
  strategy {
    type exponential_backoff
  }
}

Create Example Function

// example.baml
class Person {
  name string
  email string @description("Email address if present")
  age int?
}

function ExtractPerson(text: string) -> Person {
  client GPT4
  prompt #"
    Extract person information from:
    {{ text }}

    {{ ctx.output_format }}
  "#
}

test BasicTest {
  functions [ExtractPerson]
  args {
    text "John Smith, [email protected], 30 years old"
  }
}

Generate Client (REQUIRED)

After creating BAML files, generate the typed client:

baml-cli generate

This creates the baml_client/ directory with:

• Type definitions (Pydantic models or TypeScript interfaces)
• Client functions you import and call
• This is 100% generated - never edit these files directly