TensorTree is a zero-operations vector database leveraging an SOP transactional B-tree architecture. This innovative approach to semantic memory transforms the way knowledge is organized and retrieved by embedding categories and items as tensors. Achieving search speeds of (O(\log n)) with zero index optimization, TensorTree utilizes relativity-inspired domain centroids to warp semantic spaces, thus effectively eliminating the pigeonhole edge-case error and significantly reducing LLM hallucinations.

Key Features

• Hierarchical Organization: Unlike conventional memory systems that treat information as a flat structure, TensorTree allows users to categorize knowledge into meaningful hierarchies. This structured approach facilitates easier navigation and retrieval through path-based queries and semantic similarity matching.

• Semantic Category-Path Matching: This breakthrough feature allows users to query abstract paths like Root/Knowledge, which can map to more specific categories such as Root/Engineering. This dynamic matching relies on understanding the meanings of path components, enhancing the retrieval process.

• Efficient Vectorization: TensorTree incorporates the Vectorize API to convert content into embeddings on the fly, enabling immediate semantic comparison without a cumbersome reindexing process. This means new content can be vectorized and utilized for retrieval without delay.

• Visualizable Categories: In TensorTree, categories are not mere labels; they can be visualized and navigated through the SOP Data Manager, enhancing comprehension and exploration of relationships within the knowledge base.

Applications

TensorTree is particularly beneficial for:

• Retrieval-Augmented Generation (RAG) systems
• AI copilots and agents
• Documentation and internal knowledge search
• Domain-specific AI assistants

Getting Started

TensorTree provides a straightforward workflow for developers using Go. Creating a KnowledgeBase involves defining categories, adding items, and subsequently retrieving relevant content with an intuitive API:

kb, err := database.NewKnowledgeBase(ctx, "demo-kb", sop.DatabaseOptions{
    Type:          sop.Standalone,
    StoresFolders: []string{"./data/demo-kb"},
}, nil, nil, false)

Conclusion

TensorTree defines a new standard in semantic memory solutions, combining the ease of use for developers with the power of structured semantic search. By transforming retrieval into a guided experience based on taxonomy and meaning, it stands out as a robust option for integrating intelligent knowledge management into applications.