LinkingMem - v0.3.0

LinkingMem is a graph-native Retrieval Augmented Generation (RAG) engine designed to provide a high-performance solution for complex knowledge retrieval tasks. It combines the speed and efficiency of Rust with the flexibility of Python AI plugins, offering a unified pipeline for vector search, graph traversal, and Large Language Model (LLM) reasoning. This integration allows for fast multi-hop retrieval, efficient memory usage, and production-ready scalability, making it suitable for managing and querying large knowledge graphs. The core problem LinkingMem addresses is the fragmentation of data retrieval systems. Traditionally, combining vector search for semantic similarity with graph traversal for relational understanding requires stitching together multiple disparate tools. This approach often leads to performance bottlenecks, increased complexity, and difficulties in maintaining data consistency. LinkingMem aims to solve this by offering a single, cohesive system that seamlessly integrates these capabilities, thereby streamlining the development and deployment of advanced AI applications that require deep contextual understanding. One of the key features of LinkingMem is its tight integration of graph and vector search. It utilizes Hierarchical Navigable Small Worlds (HNSW) for efficient vector similarity search and Breadth-First Search (BFS) for graph traversal. This dual approach allows the system to not only find semantically similar information but also to explore the relationships between data points, enabling more nuanced and context-aware retrieval. The engine is built with Rust for performance and includes Python AI plugins for extensibility, allowing users to leverage various LLMs and embedding models. Another significant capability is its embedding-based entity resolution. This feature helps in accurately identifying and linking related entities within the knowledge graph, even if they are represented differently. By using embeddings, LinkingMem can resolve ambiguities and ensure that the graph accurately reflects the connections between concepts. This is crucial for maintaining the integrity and usefulness of the knowledge graph, especially when dealing with large and complex datasets. LinkingMem also offers pluggable LLM and embedding backends. This flexibility allows users to choose the best models for their specific needs, whether it's for generating embeddings, performing semantic searches, or synthesizing answers. The system's architecture is designed to accommodate different AI models without requiring significant code changes, promoting adaptability and future-proofing. For low-latency storage and fast access, LinkingMem employs memory-mapped files (mmap). This technique allows the system to access data directly from disk as if it were in memory, significantly reducing I/O overhead and improving response times. This is particularly beneficial for applications that require real-time or near real-time data retrieval, especially when dealing with large knowledge graphs. LinkingMem's overall approach is to provide a unified, high-performance engine for graph-native RAG. It achieves this by combining Rust's performance characteristics with Python's AI ecosystem. The system processes queries through a pipeline that includes embedding generation, HNSW retrieval, graph expansion via BFS, ranking, and LLM-based answer generation. This integrated pipeline ensures that vector search and graph traversal work in concert, enabling fast multi-hop reasoning and efficient resource utilization. The benefits for users include faster and more accurate retrieval of information, especially for complex queries that require understanding relationships between data points. The system's scalability ensures that it can handle large knowledge graphs, and its flexible architecture allows for easy integration with existing AI workflows. The unified nature of the engine reduces development complexity and operational overhead. Concrete use cases for LinkingMem include building advanced question-answering systems that can traverse knowledge graphs to find answers, developing recommendation engines that leverage both semantic similarity and relational data, and creating intelligent agents that can reason over complex information structures. The multimodal capabilities, supporting both text and image nodes in the same vector space, open up new possibilities for visual search and analysis. LinkingMem is positioned as a tool for developers and organizations building AI-powered applications that require sophisticated knowledge retrieval. It is available as Docker images, including an all-in-one version and an engine-only version, simplifying deployment. The project is open-source, with its code hosted on GitHub, and it is free to use. In summary, LinkingMem offers a powerful, integrated solution for graph-native RAG, combining high performance, flexibility, and scalability to enable advanced knowledge retrieval and reasoning capabilities for AI applications.