Qdrant lowers vector database complexity to aid development
The vendor now provides an integrated environment for operationalizing unstructured data to aid enterprises struggling with a profusion of tools to generate and manage vectors.
Qdrant on Tuesday launched Qdrant Cloud Inference, a managed service that enables users to generate vector embeddings for unstructured data in the same database environment in which they are stored and indexed, removing complexity and speeding application development.
Traditionally, vector embeddings -- numerical representations of unstructured data such as text and images that give them a form of structure to make them discoverable -- are generated in their own environment before being moved to another where they are stored and indexed.
By enabling users to create vector embeddings in the same database environment where they are stored and indexed, Qdrant is looking to reduce the need for users to oversee multiple vector infrastructures, manage manual pipelines for moving vector embeddings and execute data transfers.
The intended results are simplicity and speed. Developers are able to build applications, including AI tools such as assistants and agents, faster and easier than when they have to move vector embeddings between environments to develop applications informed by their organization's proprietary data.
Qdrant Cloud Inference is significant for users, as it unifies embedding generation and vector search in one environment, eliminating separate infrastructure needs and reducing latency and network costs.
Stephen CatanzanoAnalyst, Enterprise Strategy Group
Because it simplifies creating and managing vectors, Qdrant Cloud Inference is valuable for the vendor's customers, according to Stephen Catanzano, an analyst at Enterprise Strategy Group, now part of Omdia.
"Qdrant Cloud Inference is significant for users, as it unifies embedding generation and vector search in one environment, eliminating separate infrastructure needs and reducing latency and network costs," he said.
Based in Berlin, Qdrant is a vector database vendor that offers open source and managed cloud versions of its platform. Competitors include fellow vector database specialists such as Pinecone and Weaviate, as well as more broad-based data management vendors such as AWS and Oracle, which have made vector search and storage part of their database capabilities.
New capabilities
While not new, vector search and storage have gained popularity over the nearly three years since OpenAI's November 2022 launch of ChatGPT marked a significant improvement in generative AI (GenAI) technology.
Because it enables true natural language processing and lets users automate certain processes, GenAI can make workers better informed and more efficient. As a result, many enterprises have boosted their investments in GenAI development.
However, GenAI is prone to hallucinations -- incorrect and sometimes bizarre responses to user queries. The more high-quality data that GenAI tools can call upon, the more accurate and less prone to hallucinations they'll be.
Vector embeddings are a means of providing GenAI tools the requisite volume of data needed to reduce inaccuracies. Unstructured data is estimated to make up more than three-quarters of the world's data. But because it's unstructured, it's difficult to search for and discover, so it often goes unused. Vector embeddings give it structure so it can be discovered and operationalized; thus, they have surged in popularity.
Now, Qdrant is aiming not simply to enable users to generate vector embeddings to make unstructured data -- specifically, text and images -- accessible within its database but to simplify the process of generating and making discoverable such embeddings.
It's estimated that most enterprises use more than five and as many as 15 tools to generate and manage vector embeddings, according to Qdrant.
To reduce that complexity -- and expense -- and make it faster for developers to build AI tools, Qdrant Cloud Inference natively integrates embedding models for text and images with the Qdrant Cloud database rather than keeping the models in a separate environment.
Initially, supported vector embedding models include MiniLM, Sparse Lexical and Expansion Model, BEST MATCHING 25, Mixedbread Embed Large and Contrastive Language-Image Pre-training, with more models to be added over time. In addition, the managed service includes up to 5 million free tokens -- units that represent a portion of unstructured data -- per model, per month, with unlimited tokens for BM25.
Regarding the impetus for developing Qdrant Cloud Inference, customer feedback was the main motivator, according to Andre Zayarni, the vendor's co-founder and CEO.
"Early in our journey, we heard a recurring theme from our community that embedding pipelines was slowing teams down," he said.
That led to the development of a tool named FastEmbed, an open source Python library for generating and storing vectors in Qdrant's database, Zayarni continued.
"But the feedback from our community kept coming -- could we take it further, and unify embedding and retrieval in one system?" he said. "That's what led us to build Qdrant Cloud Inference."
Meanwhile, because Qdrant Cloud Inference not only unifies vector generation and indexing in one database environment but also integrates text and image embeddings, the feature is a potential differentiator for Qdrant, according to Catanzano.
While other vector databases offer unified database capabilities for embedding generation and management, they don't offer multimodal capabilities within that same environment.
"This capability appears relatively unique in the market," Catanzano said. " While not explicitly a breakthrough that will force competitors to respond immediately, it represents an important evolution in simplifying AI workflows that competitors will likely need to address. The integration of … embeddings in one managed environment is particularly notable."
Looking ahead
With Qdrant Cloud Inference now generally available, the vendor's primary focus will be on improving the performance -- specifically the efficiency and recall -- of its vector search capabilities, according to Zayarni.
Included in that is Qdrant's work with Microsoft on Natural Language Web to provide natural language interfaces for vector search.
"It's part of our broader goal to stay focused on real-time, high-precision search and unlock new ways developers can build with vectors without adding complexity," Zayarni said.
Catanzano, meanwhile, suggested several ways for Qdrant to evolve, including adding more model choices for developers, improving multimodal capabilities and developing vertical-specific tools to attract customers in specific industries.
In addition, partnering with more AI development providers would be beneficial, he continued.
"Improving integration with popular AI frameworks and expanding its hybrid cloud options would further strengthen its market position," Catanzano said.
Eric Avidon is a senior news writer for Informa TechTarget and a journalist with more than 25 years of experience. He covers analytics and data management.
