roovy-sdk 0.3.1

Python SDK for generating WhatsApp Flows and Message Templates using LLMs

Roovy SDK

A Python SDK for generating WhatsApp Flows and Message Templates using LLMs. Generate valid, schema-compliant JSON from natural language descriptions.

Features

• 🚀 Simple API - Generate WhatsApp Flows and Message Templates with a single function call
• ✅ Schema Validation - Built-in Pydantic models ensure valid output
• 🔄 Auto-Retry - Automatic retries with error feedback for better results
• 📡 Streaming Support - Stream responses for real-time feedback
• 🔌 Provider Agnostic - Works with OpenAI, Groq, Ollama, or any OpenAI-compatible API
• 📝 Type Safe - Full type hints and Pydantic models
• 📋 Dual Output - Template generation returns both creation and send payloads

Installation

pip install roovy-sdk

Quick Start

Generate a WhatsApp Flow

from roovy_sdk import RoovyClient

client = RoovyClient(
    api_key="your-api-key",
    base_url="https://api.openai.com/v1",  # or any OpenAI-compatible endpoint
    model="gpt-4o",
)

# Generate a WhatsApp Flow from natural language
flow = client.generate(
    "Create a customer feedback form with name, email, rating (1-5 stars), and comments"
)

# Access the validated flow
print(flow.model_dump_json(indent=2, by_alias=True))

Generate a Message Template

# Generate both creation + send payloads
result = client.generate_template(
    "Send John a shipping notification for order ORD-456"
)

# Template creation JSON (register with Meta for approval)
print(result.creation.model_dump_json(indent=2))

# Send message JSON (ready to send to user)
print(result.send.model_dump_json(indent=2))

Smart placeholder extraction: The SDK automatically extracts values from your prompt into the send payload. If no meaningful values are found, it uses descriptive placeholders like <CUSTOMER_NAME>.

Configuration

Using Environment Variables

import os
from roovy_sdk import RoovyClient

os.environ["ROOVY_API_KEY"] = "your-api-key"
os.environ["ROOVY_BASE_URL"] = "https://api.openai.com/v1"
os.environ["ROOVY_MODEL"] = "gpt-4o"

client = RoovyClient()

Provider Examples

OpenAI:

client = RoovyClient(
    api_key="sk-...",
    base_url="https://api.openai.com/v1",
    model="gpt-4o",
)

Groq:

client = RoovyClient(
    api_key="gsk_...",
    base_url="https://api.groq.com/openai/v1",
    model="llama-3.3-70b-versatile",
)

Ollama (local):

client = RoovyClient(
    api_key="ollama",  # any non-empty string
    base_url="http://localhost:11434/v1",
    model="llama3.2",
)

Advanced Usage

Streaming Generation

def on_chunk(chunk: str):
    print(chunk, end="", flush=True)

# Stream a flow
flow = client.generate_stream(
    "Create a booking form for a concert",
    on_chunk=on_chunk,
)

# Stream a template
result = client.generate_template_stream(
    "Create a promotional offer template with image",
    on_chunk=on_chunk,
)

Custom Retry Settings

flow = client.generate(
    "Create a survey form",
    max_retries=5,
    temperature=0.3,
)

Custom System Prompt

flow = client.generate(
    "Create a registration form",
    system_prompt="You are a WhatsApp Flows expert...",
)

Direct Schema Access

from roovy_sdk.schema import (
    WhatsAppFlow,
    Screen,
    TextInput,
    Footer,
    validate_flow,
)

# Build flows programmatically
screen = Screen(
    id="WELCOME",
    title="Welcome",
    layout=SingleColumnLayout(
        type="SingleColumnLayout",
        children=[
            TextHeading(type="TextHeading", text="Hello!"),
        ]
    )
)

# Validate any JSON against the schema
result = validate_flow({"version": "7.2", "screens": [...]})
if result["success"]:
    flow = result["data"]

Template Validation

from roovy_sdk import validate_template_result

result = validate_template_result({
    "creation": {"name": "my_template", "category": "UTILITY", "components": [...]},
    "send": {"messaging_product": "whatsapp", "to": "...", "type": "template", "template": {...}}
})
if result["success"]:
    template = result["data"]  # TemplateResult object

Schema Reference

Flow Components

The SDK includes complete Pydantic models for all WhatsApp Flow components:

Text: TextHeading, TextSubheading, TextBody, TextCaption

Input: TextInput, TextArea, Dropdown, RadioButtonsGroup, ChipsSelector, CheckboxGroup, OptIn

Date/Media: CalendarPicker, DatePicker, PhotoPicker, DocumentPicker, Image

Navigation: Footer, Button, EmbeddedLink

Conditional: If, Switch

Container: Form, SingleColumnLayout

Template Components

Result: TemplateResult (contains .creation and .send)

Creation: TemplateCreation, HeaderComponent, BodyComponent, FooterComponent, ButtonsComponent

Buttons: QuickReplyButton, UrlButton, PhoneNumberButton

Send: TemplateSend, SendComponent, TextParameter, ImageParameter, VideoParameter, DocumentParameter

Supported Template Types

Type	Category	Description
Text only	UTILITY / MARKETING	Body with {{N}} placeholders
Header Text	UTILITY	Static text header + body
Image Header	MARKETING	Image header + body
Document Header	UTILITY	Document attachment + body
Video Header	MARKETING	Video header + body
Quick Reply Buttons	UTILITY	Up to 3 quick reply buttons
URL Button	MARKETING	Up to 2 URL buttons
Call Button	UTILITY	1 phone number button
Authentication (OTP)	AUTHENTICATION	OTP verification message

API Reference

RoovyClient

RoovyClient(
    api_key: str | None = None,      # API key (or ROOVY_API_KEY env var)
    base_url: str | None = None,     # Base URL (or ROOVY_BASE_URL env var)
    model: str | None = None,        # Model name (or ROOVY_MODEL env var)
)

client.generate()

client.generate(
    prompt: str,                     # Natural language description
    *,
    max_retries: int = 3,           # Number of retry attempts
    temperature: float = 0.2,        # LLM temperature
    system_prompt: str | None = None, # Custom system prompt
) -> WhatsAppFlow

client.generate_stream()

client.generate_stream(
    prompt: str,                     # Natural language description
    on_chunk: Callable[[str], None] | None = None,  # Chunk callback
    *,
    temperature: float = 0.2,        # LLM temperature
    system_prompt: str | None = None, # Custom system prompt
) -> WhatsAppFlow

client.generate_template()

client.generate_template(
    prompt: str,                     # Natural language description
    *,
    max_retries: int = 3,           # Number of retry attempts
    temperature: float = 0.2,        # LLM temperature
    system_prompt: str | None = None, # Custom system prompt
) -> TemplateResult                  # Has .creation and .send properties

client.generate_template_stream()

client.generate_template_stream(
    prompt: str,                     # Natural language description
    on_chunk: Callable[[str], None] | None = None,  # Chunk callback
    *,
    temperature: float = 0.2,        # LLM temperature
    system_prompt: str | None = None, # Custom system prompt
) -> TemplateResult

validate_flow()

from roovy_sdk import validate_flow

result = validate_flow(flow_dict)
# Returns: {"success": True, "data": WhatsAppFlow} or {"success": False, "error": ValidationError}

validate_template_result()

from roovy_sdk import validate_template_result

result = validate_template_result(template_dict)
# Returns: {"success": True, "data": TemplateResult} or {"success": False, "error": ValidationError}

Changelog

v0.3.1 (2026-03-03)

📖 Documentation & PyPI Update

• Updated: README with full template generation documentation, API reference, and usage examples
• Added: Changelog to PyPI project page

v0.3.0 (2026-03-03)

🎉 Message Template Generation

• New: client.generate_template() — Generate WhatsApp Message Templates from natural language
• New: client.generate_template_stream() — Streaming version of template generation
• New: TemplateResult — Returns both template creation JSON (for Meta approval) and send message JSON
• New: Smart placeholder extraction — auto-extracts values from prompts into send payload parameters
• New: template_schema.py — Complete Pydantic models for all template types (Text, Image, Document, Video headers; Quick Reply, URL, Call buttons; OTP/Authentication)
• New: Built-in validation for button limits (max 3 Quick Reply, max 2 URL, max 1 Call)
• New: validate_template_result() and validate_template_creation() helpers

v0.2.3 (2026-03-03)

🔧 Terminal Screen Fix

• Fixed: Terminal screen in generated flows now shows a thank-you/confirmation message instead of a data review screen
• Fixed: System prompt updated with explicit Rule #9 — terminal screens must NOT display user data back as TextBody lines
• Fixed: Working example in system prompt corrected to use "🎉 Booking Confirmed!" pattern
• Added: "success": true to terminal screen in examples

v0.2.2 (2026-03-02)

📦 Initial Public Release

• WhatsApp Flow generation from natural language via client.generate()
• Streaming support via client.generate_stream()
• Complete Pydantic schema for all WhatsApp Flow components (v6.3–v7.2)
• Auto-retry with validation error feedback
• Provider agnostic — works with OpenAI, Groq, Ollama, or any OpenAI-compatible API
• Bundled WhatsApp Flows documentation for enhanced LLM context
• validate_flow() helper for validating arbitrary JSON

License

MIT License - see LICENSE for details.