🔥 Nuclia is now Progress Agentic RAG   Learn More

Nuclia integrations for modular RAG

Get Nuclia’s modular RAG-as-a-Service and build the perfect RAG pipeline for your specific use case.

Vector DBs

Nuclia end-to-end RAG seamlessly integrates with several vector databases, including Pinecone and our very own NucliaDB. These integrations enable powerful and efficient retrieval capabilities, allowing you to maximize the potential of your unstructured data with the best-in-class vector search technology.

Pinecone enables rapid search and retrieval of vectors, essential for applications that demand real-time data processing. Whether you’re working with large datasets or complex queries, Pinecone ensures that you get the results you need without delay.

Learn more 

NucliaDB is the open-source vector database developed by Nuclia, designed to efficiently manage and retrieve unstructured data using vector search technology. 

GitHub

 

Language models

Nuclia integrates effortlessly with a variety of leading Large Language Models (LLMs), giving you the flexibility to choose the best model for your specific needs. Whether you’re using OpenAI, Mistral, Anthropic, or other cutting-edge LLMs, Nuclia ensures seamless deployment and optimal performance across all your AI-driven projects.

OpenAI models available in Nuclia: 

  • OpenAI + Azure
  • ChatGPT-4 Turbo
  • OpenAI + Azure
  • ChatGPT-4o
  • OpenAI + Azure
  • ChatGPT-3.5
  • OpenAI ChatGPT-Vision
  • OpenAI ChatGPT-4
  • OpenAI ChatGPT-4o
  • OpenAI ChatGPT-4o-mini

Anthropic models available in Nuclia: 

  • Anthropic Claude 3
  • OpusAnthropic Claude 3 Sonet
  • Anthropic Claude 3.5 Sonet

Google Gemini models available in Nuclia:

  • Google Gemini Pro 1.5
  • Google Gemini Pro 1.5 Vision

Mistral models available in Nuclia:

  • Mistral AI Mixtral
  • Mistral + Azure Mistral large

Use Nuclia’s integration with Hugging Face to leverage any model available on Hugging Face for your own RAG. 

Learn more

Embedding models

Nuclia empowers you to take full control of your data processing by allowing you to choose the embedding model that best aligns with your specific data needs. Whether you’re dealing with text, images, or other types of unstructured data, Nuclia offers the flexibility to select from a range of advanced embedding models. 

Use Nuclia’s integration with Hugging Face to leverage any embedding model available on Hugging Face for your own RAG. 

Learn more

Nuclia embeddings

Nuclia’s proprietary embedding model is designed for optimal performance with multilingual datasets, providing accurate and efficient processing across languages.

OpenAI 3 large & small

OpenAI’s small embeddings are efficient, lower-dimensional vectors ideal for basic tasks, while large embeddings offer high-dimensional, detailed representations for complex analyses.

Google Gecko

Google’s Gecko embeddings come in small, efficient vectors for simpler tasks and large, detailed vectors for deep, complex data analysis and understanding.

Retrieval integrated strategies

Nuclia Retrieval component allows to choose the optimal retrieval strategy for specific use case. New strategies are constantly added thanks to Nuclia’s community and they keep evolving with the state-of-the-art development. Find below some examples of Nuclia’s out-of-the-box retrieval strategies.

Textual Hierarchy

This strategy prepends matched paragraphs with the title and summary of the resource, and optionally extends the paragraph with subsequent text. It’s particularly useful when paragraphs are semantically relevant but lack sufficient context.

Pass Entire Resources as Context

This strategy provides the entire resources containing matched paragraphs as context to the generative model. While it offers maximal semantic context, it may quickly reach the token limit of the model. It is best suited for small resources.

Pass Specific Field(s) as Context

This strategy appends specific fields from the resource to the matched paragraph. For example, in a Knowledge Box containing contracts, this strategy could append an “updates” field to provide additional context.

Include images

For users utilizing visual LLMs like OpenAI ChatGPT-Vision or Google Gemini Pro Vision, appending images to the context can be valuable. Nuclia offers two strategies to include images:

• Page Images: Append images present on the same page as the matched paragraph.

• Paragraph Images: Append images present next to the matched paragraph.

Add Extra Context

The extra_context parameter allows users to add a list of text snippets to the context.

  • When the Knowledge Box does not contain all the relevant information to generate an answer. For instance, if the Knowledge Box contains your product catalog but lacks user preferences, you can add the current user’s preferences as extra context.

    • When you need to run multiple distinct searches to gather all necessary information. The extra_context parameter can pass the combined results.

    • When you need to chain several questions to generate the answer. The extra_context parameter can pass the results of previous questions.

Ask a Specific Resource

By calling the /ask endpoint directly on the resource URL (rather than on the Knowledge Box URL), the RAG mechanism is bypassed, and no /find call is made. The full resource content is used as context for the generative model.