DataDesigner/docs/concepts/deployment-options.md

Deployment Options: Library vs. Microservice

Data Designer is available as both an open-source library and a NeMo Microservice. This guide helps you choose the right deployment option for your use case.

Deployment Architectures at a Glance

Data Designer supports three main deployment patterns:

• Library + Your LLM Provider

  ---

  Each user runs the library locally and connects to their choice of LLM provider.

  

• Library + Enterprise Gateway

  ---

  Users run the library locally but share a centralized enterprise LLM gateway with RBAC and governance.

  

• SDG as a Service (Microservice)

  ---

  A centralized SDG service that multiple users access via REST API.

  

Quick Comparison

Aspect	Open-Source Library	NeMo Microservice
What it is	Python package you import and run	REST API service exposing preview and create methods
Best for	Developers with LLM access who want flexibility and customization	Teams using NeMo Microservices platform
LLM Access	You provide (any OpenAI-compatible API)	Integrated with NeMo Microservices Platform
Installation	pip install data-designer	Deploy via NeMo Microservices platform
Scaling	You manage inference capacity	Managed alongside other NeMo services

!!! success "Same Configuration API" Both the library and microservice use the same DataDesignerConfigBuilder API. Start with the library, and your configurations migrate seamlessly if you later adopt the NeMo platform.

📦 When to Use the Open-Source Library

The library is the right choice for most users. Choose it if you:

You Have Access to LLMs

{ align=right width="350" }

You have API keys or endpoints for LLM inference:

• Cloud APIs: NVIDIA API Catalog (build.nvidia.com), OpenAI, Azure OpenAI, Anthropic
• Self-hosted: vLLM, TGI, TensorRT-LLM, or any OpenAI-compatible server
• Enterprise gateways: Centralized LLM gateway with RBAC, rate limiting, or other enterprise features

from data_designer.interface import DataDesigner
from data_designer.config import ModelConfig

# Use any OpenAI-compatible endpoint
model = ModelConfig(
    alias="my-model",
    model="nvidia/nemotron-3-nano-30b-a3b",
    provider="nvidia",  # or "openai", or a custom ModelProvider
)

dd = DataDesigner()
# Your code controls the full workflow

You Need Maximum Flexibility

• Custom plugins: Extend Data Designer with custom column generators, seed readers, or processors
• Local development: Rapid iteration with immediate feedback
• Integration: Embed Data Designer into existing Python pipelines or notebooks
• Experimentation: Research workflows with custom models or configurations

You Already Have Enterprise LLM Infrastructure

{ align=right width="350" }

!!! tip "Library + Enterprise LLM Gateway" Many enterprises already have centralized LLM access through API gateways with:

- Role-based access control (RBAC)
- Rate limiting and quotas
- Audit logging
- Cost allocation

In this case, **use the library** and point it at your enterprise gateway. You get enterprise-grade LLM access while retaining full control over your Data Designer workflows.

from data_designer.config import ModelConfig, ModelProvider

# Define your enterprise gateway as a provider
enterprise_provider = ModelProvider(
    name="enterprise-gateway",
    endpoint="https://llm-gateway.yourcompany.com/v1",
    api_key="ENTERPRISE_LLM_KEY",  # Environment variable name (uppercase) or actual key
)

# Use the provider in your model config
model = ModelConfig(
    alias="enterprise-llm",
    model="gpt-4",
    provider="enterprise-gateway",  # References the provider above
)

---

☁️ When to Use the Microservice

{ align=right width="350" }

The NeMo Microservice exposes Data Designer's preview and create methods as REST API endpoints. Choose it if you:

You're Using the NeMo Microservices Platform

The primary value of the microservice is integration with other NeMo Microservices:

• NeMo Inference Microservices (NIMs): Seamless integration with NVIDIA's optimized inference endpoints
• NeMo Customizer: Generate synthetic data for model fine-tuning workflows
• NeMo Evaluator: Create evaluation datasets alongside model assessment
• Unified deployment: Single platform for your entire AI pipeline

You Want to Expose SDG as a Team Service

If you need to provide synthetic data generation as a shared service:

• Multi-tenant access: Multiple teams submit generation jobs via API
• Job management: Queue, monitor, and manage generation jobs centrally
• Resource sharing: Shared infrastructure for SDG workloads

When users can submit configs containing Jinja templates to a shared engine, template rendering becomes a remote code execution concern and part of your security boundary. See Security for guidance on when to keep the default JinjaRenderingEngine.SECURE mode.

---

🧭 Decision Flowchart

                    ┌─────────────────────────┐
                    │ Are you using the NeMo  │
                    │ Microservices platform? │
                    └───────────┬─────────────┘
                                │
                    ┌───────────┴───────────┐
                    ▼                       ▼
                   YES                      NO
                    │                       │
                    ▼                       ▼
        ┌───────────────────┐   ┌───────────────────────────┐
        │ Use Microservice  │   │ Do you need to expose SDG │
        │                   │   │ as a shared REST service? │
        │ Integrates with   │   └─────────────┬─────────────┘
        │ NIMs, Customizer, │                 │
        │ Evaluator         │     ┌───────────┴───────────┐
        └───────────────────┘     ▼                       ▼
                                 YES                      NO
                                  │                       │
                                  ▼                       ▼
                      ┌─────────────────────┐   ┌─────────────────┐
                      │ Consider if the     │   │ Use the Library │
                      │ overhead is worth   │   │                 │
                      │ it vs. library +    │   │ Most flexible   │
                      │ enterprise gateway  │   │ option for      │
                      └─────────────────────┘   │ direct use      │
                                                └─────────────────┘

---

Learn More

• Library: Continue with this documentation
• Microservice: See the NeMo Data Designer Microservice documentation{target="_blank"}
• Security model: See Security