Graphwise Bolsters GenAI App Development with GraphDB Update

Graph databases are emerging as a key piece of the data stack for developing and running generative AI applications. To that end, Graphwise today took the wraps off a new version of its GraphDB database that will help customers with their GenAI projects.

GraphDB is a semantic graph database designed to store data in triples (or tuples) and serve data via graph queries written in RDF and SPARQL standards. The database, which was originally developed by a company called Ontotext, historically has been used to create knowledge stores that allow organizations to store and query data about people, places, and things.

Ontotext and Semantic Web Company merged in October 2024 and created a new company called Graphwise. The merger led to the integration of GraphDB with PoolParty, the name of Semantic Web Company’s knowledge and content management offerings.

GraphDB version 11 is the first major release of the database following the merger and creation of Graphwise, and it brings a range of AI capabilities, including broader support for LLMs, enhanced GraphRAG, and the addition of MCP for agentic AI.

Using a graph database with retrieval-augmented generation (RAG), or GraphRAG as it’s known, is giving customers better answers from LLMs with fewer hallucinations. With GraphDB 11, customers will find that the database’s “Talk to Your Graph” feature now allows them to utilize a range of large language models (LLMs), including Qwen, Llama, Gemini, DeepSeek, and Mistral for natural language queries. Customers can also query their graph database using local and custom AI models.

Graphwise says that enterprise knowledge graphs like GraphDB allow LLMs to retrieve precise, context-rich information from structured data, as well as to filter out irrelevant information.

(Daniel Chetroni/Shuttersetock)

“Our vision is that, while LLMs can do all sorts of magic in terms of transformation of text and data, RAG results will only be fit for purpose if appropriate data is provided,” Graphwise President Atanas Kiryakov tells BigDATAwire via email says. “Precise RAG requires precise retrieval. Feeding LLM with inappropriate documents and data will always result in inaccurate and irrelevant results. Precise retrieval requires quality data and metadata. Vector databases cannot do this alone. Graph databases are imperative to filter out inappropriate data.”

GraphDB 11 also introduces a new entity linking service that helps to map phrases in natural language to the correct concepts or entities in the knowledge graph, thereby eliminating ambiguity and ensures that information retrieval is both precise and relevant. Both Ontotext and Semantic Web Company had created various entity linking models over the years, and with GraphDB 11, those models are now available as additional retrieval methods in the GraphRAG solution.

“This allows AI engineers to choose the right option, based on tradeoffs across speed, cost and accuracy, and on specific domain and service level agreements (SLAs),” Kiryakov says. “Our entity linking models deliver better accuracy and cost compared to using a combination of vector databases and an LLM for this purpose. GraphDB’s GraphRAG capabilities ensure outputs are not only fast, but precise, relevant, and grounded in an organization’s data as well as pre-existing ontologies and domain models.”

Graph databases offer a potentially superior solution for RAG workloads thanks to the way the data is modeled, stored, and queried, Kiryakov says. In particular, the advantages stem from their use of Web Ontology Language (OWL) ontologies, Simple Knowledge Organization System (SKOS) taxonomies, and Shapes Constraint Language (SHACL), he says.

These structures enable the graph database to “act as an extensive machine-readable documentation of the structure and the meaning of the data,” Kiryakov continues. “LLMs have already been trained on these knowledge representation languages, so, using these semantic models is a compact and unambiguous method to instruct an LLM with domain knowledge, as compared to free text prompting.”

GraphDB 11 also supports Model Context Protocol (MCP), which Anthropic launched last year to help connect AI agents to data sources. GraphDB brings other new features, such as support for GraphQL as another way to query data in the database. While RDF and SPARQL remain powerful for complex semantic queries and data modeling, adding GraphQL allows Graphwise to lower the barrier of entry for developers, Kiryakov says.

This release also brings a new ooptimized cache management functionality, which will minimize memory overhead associated with hosting multiple repositories, thereby reducing the cost and boost the efficiency of enterprise deployments, the company says.

The Future of GenAI: How GraphRAG Enhances LLM Accuracy and Powers Better Decision-Making

What’s the Vector, Victor?

The post Graphwise Bolsters GenAI App Development with GraphDB Update appeared first on BigDATAwire.