Google Cloud Next releases AlloyDB AI preview version, empowering developers to build generative AI applications.

2023-09-07

Google announced the preview version of AlloyDB AI as an important component of AlloyDB for PostgreSQL during the recent Google Cloud Next event. This allows developers to build generative artificial intelligence (AI) applications using large language models (LLMs) and real-time operational data with the help of built-in end-to-end vector embedding support.

Earlier, the company introduced pgvector support for Cloud SQL for PostgreSQL and AlloyDB for PostgreSQL, providing vector search operations for managed databases, enabling developers to store vector embeddings generated by large language models (LLMs) and perform similarity searches. AlloyDB AI builds on the basic vector support provided by standard PostgreSQL. According to the company, developers can "create and query embeddings, find related data with just a few lines of SQL code - without the need for a specialized data stack or data migration."

In addition, AlloyDB AI brings some other new features to AlloyDB that can help developers integrate their real-time data into generative AI applications:

  • Enhanced vector support with faster query speeds compared to standard PostgreSQL, thanks to tight integration with the AlloyDB query processing engine. The company has also introduced quantization technology based on its ScaNN technology, supporting four times the vector dimensions and three times the space reduction.
  • Access to locally stored models in AlloyDB and models hosted in Vertex AI, including custom and pre-trained models. Developers can train and fine-tune models using data stored in AlloyDB and then deploy them as endpoints on Vertex AI.
  • Integration with the AI ecosystem, including Vertex AI Extensions (to be launched later this year) and LangChain, which will provide the ability to call remote models in Vertex AI using SQL for low-latency, high-throughput use cases such as fraud detection.

In a blog post, Andi Gutmans, General Manager and Vice President of Engineering at Google Cloud's database division, wrote:

"AlloyDB AI allows users to easily transform their data into vector embeddings using simple SQL functions, and the speed of vector queries is 10 times faster than standard PostgreSQL. The integration with the open-source AI ecosystem and Google Cloud's Vertex AI platform provides an end-to-end solution for building generative AI applications."

In response to a question on a Reddit post about whether Google is trying to embrace, extend, and extinguish (EEE) PostgreSQL with AlloyDB AI, one respondent said:

"I think what you're trying to say is that just because someone - especially [a big company] - tries to improve/integrate a popular open project doesn't mean it's always EEE."


"I doubt EEE is mostly intentional at first, even if it may become that later. For Google, I think it's just a case of 'how do we add value to our product to sell it' and then 'this feature is too costly for us to maintain, let's cut it and focus on [new features].'"

Furthermore, other databases and public cloud providers already support vector embeddings, including MongoDB, DataStax's Cassandra database service Astra, open-source PostgreSQL (via Pgvector), and Azure Cognitive Search. The latter recently launched a preview version of a new feature that allows indexing, storage, and retrieval of vector embeddings in search indexes.

Finally, AlloyDB AI is available at no additional cost in Google Cloud and AlloyDB Omni on AlloyDB.

RECENT AI NEWS

RECENT AI TOOLS