September 2025 Summaries

SingleStore's ML Functions offer a platform for building, deploying, and serving custom machine learning models directly within the database environment, enhancing businesses' ability to leverage their unique data for predictive tasks. Unlike AI Functions, which utilize pre-trained models for general tasks like translation and summarization, ML Functions allow for the creation of tailored models that can address specific business needs, such as fraud detection, demand forecasting, and anomaly detection, by learning from the organization's own data patterns. The system supports a "No-ETL, No-Notebook" workflow through a guided user interface that simplifies model training into a single SQL command, ensuring high performance and real-time predictions by co-locating models and data. Furthermore, SingleStore integrates a comprehensive MLOps framework to monitor model health across various metrics, such as data integrity and predictive performance, to preemptively address potential degradation in model accuracy. As organizations strive for faster insights and real-time decision-making, ML Functions remove traditional barriers between data and machine learning teams, offering seamless integration with existing SQL-based workflows and aiming to democratize machine learning by making it more accessible and adaptable to changing business requirements.

SingleStore has introduced Python User-Defined Functions (UDFs) to allow the execution of Python code directly within the database, enhancing the integration of data storage with advanced analytics. This feature enables users to define custom functions in Python that can be called from SQL queries, leveraging Python's extensive data processing and machine learning libraries while maintaining SingleStore's security, performance, and scalability. The integration addresses key challenges in data architecture by eliminating data movement, bridging the gap between SQL and Python, and simplifying data pipelines, thus empowering data scientists to work with database data without needing to learn complex SQL. Python UDFs support both scalar and vectorized operations for optimized data processing, and they facilitate advanced analytics such as machine learning integration and text analytics directly within the database. The upcoming release will include support for common Python data science libraries, and further resources such as documentation and tutorials will be provided to assist users in utilizing this new capability effectively.

SingleStore has introduced AI Functions, a new capability that allows users to execute generative AI and embedding models directly on their data using SQL, thereby eliminating the need for data movement and reducing latency, complexity, and cost. This feature enables business analysts and application engineers to perform tasks like sentiment analysis, translation, and text summarization using SQL, without requiring new frameworks or complex APIs. AI Functions are designed to operate in real-time due to SingleStore's Hybrid Transactional/Analytical Processing (HTAP) architecture, which allows them to run on live data instantly. These functions, delivered as managed Python User-Defined Functions within SingleStore Aura cloud service, include tools like AI_COMPLETE for text generation and AI_SENTIMENT for sentiment analysis. By leveraging a containerized approach, AI Functions aim to rapidly adapt to AI innovation, offering a comprehensive toolkit for natural language processing and vector embedding tasks. Additionally, the architecture's massively parallel processing and HTAP design enhance performance, enabling operational AI applications like real-time fraud detection and customer support automation. SingleStore plans to expand this capability with more AI and machine learning functions, including time-series forecasting and anomaly detection, to further support enterprise operational AI needs.

In late 2024, SingleStore acquired BryteFlow, integrating its no-code, user-friendly data transfer capabilities to create SingleStore Flow, a tool designed to facilitate seamless data migration from popular databases such as MySQL, Postgres, Oracle, SQL Server, and Snowflake into SingleStore without requiring any coding. SingleStore Flow supports continuous Change Data Capture (CDC) for real-time analytics, except for Snowflake, which it handles through one-time parallel loads. This tool significantly reduces the labor and timelines associated with traditional data migration methods by enabling faster and more efficient data transfers, making it cost-effective compared to other data transfer and CDC tools. Since December 2025, Flow has been available as a fully-managed service on SingleStore Helios®, further simplifying the data transfer process by eliminating the need for users to manage hardware or software. SingleStore is committed to supporting existing BryteFlow customers and plans to enhance Flow by expanding its database compatibility, integrating its services, and incorporating AI-enabled setup and control.

SingleStore Aura has introduced a credit-based pricing model for container resources, designed to offer development teams greater flexibility in building and scaling data and AI applications. This update, still in preview, will expand to all standard and enterprise customers shortly, addressing the issue of fixed resource limits that can lead to over-provisioning or inefficiency. The platform allows users to select from small, medium, and GPU container types tailored for different workloads, such as data exploration, machine learning, and real-time analytics, enabling dynamic scaling based on actual demands. SingleStore Aura also implements safeguards like limits on active container sessions and role-based access controls to manage resource use effectively. Administrators can manage permissions and monitor usage, with future plans to enhance visualizations and role-specific access configurations. Users are encouraged to optimize container usage by assessing workload requirements and tracking resource consumption through the Aura Compute sessions page.

SingleStore has announced a growth buyout agreement with Vector Capital, a private equity firm with a strong track record in scaling software companies, marking an acceleration of its ongoing journey. The partnership will not alter SingleStore's mission or leadership but will provide additional resources to enhance its focus on innovation and customer commitment. With a recent successful quarter, including $123 million in revenue and a 47% increase in cloud growth, the investment aims to further drive growth and product innovation, particularly in AI and next-generation workloads. Vector Capital's investment is driven by confidence in SingleStore's technology and its customer loyalty, positioning SingleStore to leverage AI's transformative potential. This partnership signifies a long-term commitment from Vector Capital to help build a multi-generational company, promising more innovation and momentum for SingleStore's customers.

Enterprises today face challenges in meeting real-time data demands due to the limitations of existing data architectures, which were primarily built for batch analytics rather than real-time experiences. While data warehouses, lakes, and the emerging lakehouses offer structured analytics, flexible storage, and unified capabilities respectively, they fall short in providing real-time responsiveness, necessary for modern digital experiences and AI systems. Gartner's report suggests combining these architectures for varied analytics needs, yet it overlooks the critical aspect of real-time performance, including streaming data ingestion and low-latency querying. SingleStore offers a solution by acting as a performance layer that complements existing architectures like Snowflake and Databricks, enabling real-time queries and AI-driven applications without compromising scalability. This approach bridges the gap between data at rest and real-time intelligence, ensuring that businesses can make informed decisions with up-to-the-moment data, without replacing their current systems.

Choosing an AI database like SingleStore involves more than just benchmarks; true value comes from building and experimenting with it in real-world scenarios. SingleStore supports innovative features such as vector search, hybrid analytics, and managing the lifecycle of large language models (LLMs) in production-like workflows. It enables semantic and vector search by allowing embeddings to be stored in vector columns and retrieved with vector similarity in SQL, aiding in tasks like real-time recommendations and personalization by blending vector similarity with transactional and behavioral features. SingleStore also supports multimodal applications that handle text, audio, and images by layering retrieval over both structured and unstructured data, which can be particularly useful for applications such as virtual assistants or language translation. The platform facilitates the evaluation and fine-tuning of LLMs by offering a notebook-driven workflow for running repeatable tests, helping refine model accuracy and response style. SingleStore's ability to ingest and transform data quickly, persist vectors in a native type, and maintain a consistent data model when moving from a notebook to a service makes it an appealing choice for AI workloads, enabling various use cases like fraud detection, autonomous systems, and real-time machine learning integration.

Retrieval-Augmented Generation (RAG) is a framework designed to enhance the accuracy and specificity of language model outputs by grounding them in relevant, up-to-date documents, thereby reducing hallucinations and improving reliability. It works by retrieving context from a knowledge source, typically stored as vectors in a vector database, and using this context to generate answers with a language model. The tutorial explains RAG's stages—retrieve, augment, and generate—and emphasizes the importance of retrieval quality, which is influenced by document chunking, embedding choices, and query strategies. The guide also explores practical implementations using Python with LangChain, OpenAI, and SingleStore, highlighting the advantages of dynamic, contextually grounded responses over static FAQ bots. Additionally, the article discusses real-time RAG applications using SingleStore and Vercel for keeping knowledge bases continuously updated, as well as agentic RAG systems that combine SQL and vector searches for enhanced retrieval strategies. These applications demonstrate the operationalization of freshness and the integration of multiple data types within a single platform to optimize retrieval efficiency and response accuracy.

Model Context Protocol (MCP) introduces a standardized method for AI models to interact with external tools, APIs, and databases, enhancing AI capabilities beyond static prompts. Elicitation, an extension of MCP, formalizes the process by allowing MCP servers to specify necessary variables and their types, transforming AI interactions into a more collaborative dialogue akin to human communication. This enhances the AI's ability to manage ambiguous queries, handle complex workflows, confirm resource-intensive operations, and prevent errors in sensitive actions by requiring explicit confirmations. While elicitation is not universally available yet, it can currently be explored through environments like Visual Studio Code, particularly with SingleStore's MCP server, which exemplifies how this approach can enhance database query interactions by refining and clarifying user requests for more accurate and relevant insights. This shift towards smarter conversations represents a significant step forward in the evolution of AI, emphasizing adaptive and context-aware interactions.

Integrating SingleStore with Django involves a detailed process of installation, configuration, and handling known limitations, using a Polls app as a practical example. This guide provides step-by-step instructions for connecting Django 4.2 to a SingleStore cluster, adjusting settings, managing migrations, and modifying models to align with SingleStore's distributed architecture. It covers the installation of the SingleStore Django backend from PyPI or source, setting up a new Django project, and configuring database settings to establish a connection to the SingleStore cluster. The tutorial addresses compatibility issues, such as setting environment variables for Django's core apps and configuring storage types to accommodate unique constraints. It also includes instructions on creating models, applying migrations, and handling many-to-many relationships, with specific attention to avoiding errors during migrations due to SingleStore's implicit DDL statement commits. The guide encourages further exploration of the django-singlestore documentation for advanced configurations and suggests testing Django applications on SingleStore's real-time analytics engine for enhanced performance and scalability.