Skip to main content

​
Introduction

ScrapeGraphAI provides custom model classes that wrap various AI services and extend functionality for specific use cases. These models integrate seamlessly with the graph-based scraping pipeline, enabling advanced features like image-to-text conversion, text-to-speech generation, and support for multiple LLM providers.

​
Available Models

​
LLM Providers

ScrapeGraphAI includes wrapper classes for several LLM providers that use OpenAI-compatible APIs:
  • DeepSeek - DeepSeek language models with OpenAI-compatible API
  • XAI - xAI Grok models with OpenAI-compatible API
  • Nvidia - NVIDIA AI Foundation models
  • OneApi - Generic OpenAI-compatible API wrapper
  • CLoD - CLōD language models with OpenAI-compatible API

​
Specialized Models

​
Model Integration

All custom models integrate with ScrapeGraphAI’s graph-based architecture. Models are configured through the graph’s config dictionary and can be used in nodes throughout the scraping pipeline.

​
Basic Usage Pattern

from scrapegraphai.graphs import SmartScraperGraph
from scrapegraphai.models import DeepSeek

# Configure the model
graph_config = {
    "llm": {
        "model": "deepseek-chat",
        "api_key": "your-api-key",
        "temperature": 0.7
    }
}

# Use in a graph
scraper = SmartScraperGraph(
    prompt="Extract the main content",
    source="https://example.com",
    config=graph_config
)

​
Direct Model Usage

You can also instantiate and use models directly: 
from scrapegraphai.models import DeepSeek, OpenAIImageToText

# LLM model
llm = DeepSeek(
    model="deepseek-chat",
    api_key="your-api-key",
    temperature=0.7
)

# Image-to-text model
itt_model = OpenAIImageToText({
    "model": "gpt-4-vision-preview",
    "api_key": "your-openai-key"
})

image_description = itt_model.run("https://example.com/image.jpg")

​
Configuration

​
LLM Configuration

LLM wrapper models (DeepSeek, XAI, Nvidia, etc.) accept standard LangChain ChatOpenAI parameters:
​
model
string
required
The model identifier to use (e.g., “deepseek-chat”, “grok-beta”)
​
api_key
string
required
API key for authentication (automatically mapped to provider-specific key)
​
temperature
float
default:"0.7"
Controls randomness in model outputs (0.0 to 1.0)
​
max_tokens
int
Maximum number of tokens to generate in responses
​
streaming
bool
default:"false"
Enable streaming responses

​
Specialized Model Configuration

Specialized models like OpenAIImageToText and OpenAITextToSpeech have their own configuration requirements. See individual model pages for details.

​
Architecture

​
Wrapper Pattern

Most LLM models use a wrapper pattern that:
  1. Extends LangChain’s ChatOpenAI class
  2. Automatically maps api_key to provider-specific parameter names
  3. Sets provider-specific base URLs
  4. Inherits all LangChain functionality
This design allows seamless integration with LangChain’s ecosystem while providing provider-specific configurations.

​
Lazy Loading

Some models (like Nvidia) use lazy loading to avoid import errors when optional dependencies aren’t installed: 
class Nvidia:
    def __new__(cls, **llm_config):
        try:
            from langchain_nvidia_ai_endpoints import ChatNVIDIA
        except ImportError:
            raise ImportError("Please install langchain-nvidia-ai-endpoints")
        return ChatNVIDIA(**llm_config)

​
Graph Integration Examples

​
Image-to-Text in OmniScraperGraph

from scrapegraphai.graphs import OmniScraperGraph

graph_config = {
    "llm": {
        "model": "gpt-4-vision-preview",
        "api_key": "your-openai-key"
    },
    "max_images": 5
}

omni_scraper = OmniScraperGraph(
    prompt="Describe all product images on this page",
    source="https://example.com/products",
    config=graph_config
)

result = omni_scraper.run()

​
Text-to-Speech in SpeechGraph

from scrapegraphai.graphs import SpeechGraph

graph_config = {
    "llm": {
        "model": "gpt-3.5-turbo",
        "api_key": "your-openai-key"
    },
    "tts_model": {
        "api_key": "your-openai-key",
        "model": "tts-1",
        "voice": "alloy"
    },
    "output_path": "summary.mp3"
}

speech_graph = SpeechGraph(
    prompt="Summarize the main attractions",
    source="https://en.wikipedia.org/wiki/Chioggia",
    config=graph_config
)

speech_graph.run()

​
Next Steps

OpenAI Models

Image-to-text and text-to-speech models

DeepSeek

DeepSeek language models integration

NVIDIA

NVIDIA AI Foundation models

xAI

xAI Grok models integration

Build docs developers (and LLMs) love

Get started for freeTalk to us