Superlinked Framework

Product · Developer tool by Superlinked

Build multi-attribute search and recommendation systems with a mixture of encoders.

An open-source Python SDK for building multi-attribute search and recommendation systems. It combines structured and unstructured data into unified embeddings for advanced RAG and agent retrieval.

Languages
English

Category: Machine Learning

Programming languages
Python
License
Open source
Deployment
Self-hosted, Cloud, Local
Programming Language
Python
Core Architecture
Mixture of Encoders
Supported Data Types
Text, Numerical, Categorical, Temporal, Spatial, Behavioral
  • Mixture-of-encoders for multi-attribute search
  • Unified embedding space for structured and unstructured data
  • Native LlamaIndex and LangChain integrations
  • Open-source Python SDK for complex retrieval logic

Overview

Superlinked Framework is an open-source Python SDK designed for building advanced multi-attribute search and recommendation systems. It enables developers to move beyond simple text-only embeddings by using a mixture-of-encoders approach that combines unstructured text with structured data such as categories, numbers, and timestamps into a unified embedding space. The framework serves as a high-level logic layer that sits on top of the Superlinked Inference Engine (SIE), providing the tools needed for complex retrieval strategies in RAG and agentic workflows. It supports real-time streaming ingestion and schema-aware indexing, allowing systems to handle multi-faceted queries with higher precision. Superlinked Framework integrates natively with major vector databases like Qdrant and Chroma, as well as orchestration tools like LlamaIndex and LangChain.

Search and Retrieval Logic

  • Mixture of Encoders: The framework enables a mixture-of-encoders approach that combines separate encoders for text, numerical values, categorical metadata, and temporal signals into a unified embedding space.
  • Multi-attribute Search: Developers can build retrieval systems that handle complex queries including temporal, spatial, behavioral, or structured data alongside unstructured text.
  • Real-time Ingestion: The framework supports schema-aware indexing and streaming ingestion to handle real-time context rather than static pre-indexed documents.
  • Framework Integrations: The library provides native integrations with popular AI orchestration frameworks including LlamaIndex and LangChain.

Developer Tools and Integration

  • Python SDK: The framework serves as a higher-level Python SDK that sits on top of the Superlinked Inference Engine (SIE) layer.
  • Open Source Library: It is released as an open-source library, with the core code available on GitHub.
  • Vector Database Support: The framework integrates with major vector databases such as Qdrant, Chroma, Weaviate, and LanceDB.
  • RAG Strategy Evaluation: Developers can use the framework to implement and evaluate different retrieval strategies for RAG pipelines.

Use cases

  • Engineers build e-commerce search engines that combine product descriptions with structured filters like price and brand.
  • AI teams implement complex RAG pipelines that retrieve documents based on both semantic meaning and temporal recency.
  • Developers create recommendation systems that factor in user behavioral data alongside content embeddings.
  • Data teams use the framework to unify multi-modal signals into a single searchable vector index.

Specifications

Programming languages
Python
Integrations
LlamaIndex, LangChain, Qdrant, Chroma, Weaviate, LanceDB, CrewAI, DSPy, Haystack
License
Open source
Deployment
Self-hosted, Cloud, Local
Programming Language
Python
Core Architecture
Mixture of Encoders
Supported Data Types
Text, Numerical, Categorical, Temporal, Spatial, Behavioral
Inference Layer
Superlinked Inference Engine (SIE)

How it works

  1. Define Schema: Developers define a schema that includes both unstructured text and structured attributes like categories or timestamps.
  2. Configure Encoders: The framework uses a mixture of encoders to transform multi-attribute data into a unified vector representation.
  3. Ingest and Index: The system ingests data and stores it in a supported vector database like Qdrant or Chroma.
  4. Execute Retrieval: Applications perform complex queries that combine semantic meaning with structured filters and weights.

FAQ

Is the framework open source?

Yes, the Superlinked library is open source and available on GitHub.

How does it handle structured data in search?

The framework uses a mixture-of-encoders approach to combine structured data like numbers, categories, and timestamps with unstructured text into a unified embedding space.

Which vector databases does it support?

It integrates with major vector databases and frameworks including LlamaIndex, LangChain, Chroma, Qdrant, and Weaviate.

Added by AntSpace from public sources (superlinked.com).

Related

More products by Superlinked