Time scale intends to further expand its open-source database platform of the same name with recent artificial intelligence capabilities announced today.
Timescale was founded in 2017 as a time series database (TSDB) technology based on the open-source PostgreSQL relational database. The combination of time series data and vectors has real value for enterprises because it helps enable generative applications of artificial intelligence using Augmented Generation (RAG) technology. That’s why Timescale has especially expanded its vector capabilities this 12 months. In June, the company announced pgvectorscale and pgai, which integrate advanced vector database capabilities with the Timescale database platform. Now Timescale goes a step further with the recent pgai Vectorizer development tool, which creates and synchronizes embeddings directly in the database. As an open source technology, the pgai vectorizer can potentially be used by any PostgreSQL database user to enable the creation of generative AI applications.
“We took this small idea for PostgreSQL into time series and then it evolved into a much bigger idea based on our success in this space, which is that PostgreSQL is a development platform for any application,” Ajay Kulkarni, CEO and co-founder of Timescale told VentureBeat.
The intersection of time series data and vector database technology
Timescale’s area of interest is the intersection of time series data and vector database technologies.
Kulkarni explained that these two types of data overlap and might be used together in different applications. He noted that customers currently using Timescale have customers who only use the database for time series, and some who only use it for vectors. The third category is customers who are beginning to use technology in each cases. The intersection of time series and vector data allows for use cases that leverage each the temporal aspect of time series and the semantic capabilities of vector mining.
Timescale’s early customers include electric vehicle startup Lucid Motors. Kulkarni explained that Lucid uses vector search for images, which also have a timestamp, and the value of images decreases over time.
Kulkarni said he sees combining time series and vector data as an vital trend, with organizations seeking to leverage the strengths of each types of data inside a single database platform similar to PostgreSQL.
The goal is to simplify the management of vector databases for artificial intelligence
The recent pgai vectorizer is an extension of Timescale’s pgai initiative, which launched in June. The initial part of this effort allows Timescale users to integrate the AI model directly with PostgreSQL.
The recent pgai vectorizer goals to streamline embedding management, making it so simple as traditional database operations. The open-source tool allows developers to create and manage embeddings across multiple text columns using easy SQL commands, mechanically staying in sync as underlying data changes. It also facilitates easy testing and deployment of different AI models, including switching between services.
Pgai Vectorizer is based on Timescale’s existing vector database technologies, launched in June 2024. The company’s pgvectorscale extension is based on the open source extension of the pgvector vector database. Many vendors, including AWS and Google, use pgvector to offer PostgreSQL with vector database capabilities
Timescale sees pgvector as having larger scale limitations that pgvectorscale is intended to deal with. According to Kulkarni, pgvectorscale provides higher performance and scalability in comparison with pgvector while remaining fully compatible and open source. He also argued that the open-source pgvectorscale could outperform other vector database technologies, including Pinecone.
Moving beyond RAG to agent-based AI for vector database operations
Kulkarni emphasized that pgai Vectorizer, like the pgvectorscale extension, is open source and will remain so. He hopes that keeping the technology open source will help grow the community of users and contributions.
Looking ahead, the company sees pgai Vectorizer as part of a broader artificial intelligence strategy.
“We are essentially building RAG as a service directly into the database,” he said. “But we don’t stop at RAG, we look at agents.”