July 2026 Summaries

ClickHouse emphasizes speed as a core feature, influencing its design and engineering, which is evident in its development of benchmarks such as ClickBench and PostgresBench. The introduction of pg_re2, a Postgres extension utilizing RE2 for regular expressions, enhances performance and compatibility with ClickHouse, offering significant speed improvements over native Postgres functions. RE2, based on finite automata rather than Postgres's backtracking POSIX regular expressions, provides more consistent execution times and high-level analyses. Benchmarks demonstrate that pg_re2 consistently outperforms Postgres in regex operations, with speed increases ranging from 1.1x to 8.6x. The extension also supports btree and GIN indexing, reducing the need for table scans. While RE2 differs from Postgres POSIX in terms of speed and syntax, it ensures compatibility with ClickHouse, facilitating seamless integration. This compatibility is crucial for analytical queries, particularly when migrating data to ClickHouse, as it maintains consistent performance and pushdown. ClickHouse Managed Postgres, available on ClickHouse Cloud, integrates these features, offering a unified data stack for scalable applications.

AI-driven applications are transforming database demands by necessitating real-time transactional and analytical processing, which has led to a growing reliance on Postgres and ClickHouse as a complementary pair of databases. This shift is driven by AI's need to handle massive data volumes and provide immediate insights, requiring a seamless integration between OLTP and OLAP systems. The open-source nature of Postgres and ClickHouse, coupled with their strong communities and proven capabilities, makes them increasingly popular among AI-native companies and enterprises. These databases are being enhanced with tools like PeerDB for Change Data Capture and the pg_clickhouse extension to ensure rapid data replication and efficient query handling. The partnership between Postgres and ClickHouse exemplifies a best-of-breed approach, offering a unified and robust data stack that meets the high concurrency and low-latency demands of modern AI applications, while avoiding the pitfalls of proprietary platform lock-ins.

chDB addresses the inefficiencies in chatbot and AI agent systems by integrating a ClickHouse query engine within the agent's process, reducing reliance on network calls and thus minimizing latency and instability. By localizing data access, chDB allows for fast, deterministic queries that bypass the compounded latency and potential failures of network-dependent systems, making it an ideal solution for maintaining a stable and efficient agent operation. It supports a comprehensive memory model that preserves data history and facilitates efficient recall, leveraging ClickHouse's capabilities for structured data, time-series, and vector storage. Additionally, chDB functions as a data federation hub, allowing seamless integration and querying across multiple data sources without the overhead of complex data-plumbing code. This setup not only enhances stability and optimizes token usage by minimizing retries and detours but also aligns with AWS Lambda MicroVMs for isolated, efficient, and scalable processing environments. The architecture ultimately seeks to improve the reliability and performance of AI agents by ensuring that critical data remains local and readily accessible, reducing the need for costly and unstable remote data calls.

ClickStack introduces AI dashboard generation, an innovative feature designed to simplify the creation of dashboards by transforming the process into an investigative workflow rather than just producing standalone dashboards. Users can now describe their desired insights, and ClickStack explores telemetry data, generates queries, and assembles visualizations within an AI Notebook, creating a collaborative workspace that captures the entire investigative process. This approach not only makes dashboard creation more intuitive and accessible, especially for new users unfamiliar with schema discovery and query writing, but also offers experienced users a more efficient means to build and refine dashboards. The generated dashboards are fully editable, allowing users to modify queries and visualizations, with the AI Notebook serving as a complete, editable history of the investigation. Beyond individual dashboards, ClickStack supports the creation of interconnected investigative workflows by linking multiple dashboards, enabling more comprehensive problem-solving capabilities. Additionally, ClickStack's MCP server enables seamless integration with external tools like Claude, Cursor, and Codex, allowing teams to leverage the same high-level observability and dashboard management tools, thus fostering a versatile and expandable observability platform.

Verihubs, an AI company specializing in identity infrastructure for major Indonesian banks and fintechs, transitioned its data warehouse from a Postgres-based setup to a ClickHouse architecture to enhance its analytics capabilities. The previous system relied on daily batch data pulls from Postgres, which led to slow query processing and challenges in data accuracy and timeliness. By adopting ClickHouse and utilizing Kafka for real-time data streaming, Verihubs achieved up to 98% faster query speeds and a 50% reduction in cloud costs, transforming its dashboards into interactive OLAP experiences. This new setup allows real-time ingestion and post-processing, significantly improving data freshness and reliability, which is crucial for the company's large-scale operations involving around 50 million API calls monthly. The migration also taught Verihubs the importance of careful database design and the nuances of ClickHouse's performance profile, particularly regarding its MergeTree engines and the handling of updates and backfills. The shift to ClickHouse not only accelerated query performance but also optimized costs and improved user experience, enabling stakeholders to access up-to-date information more efficiently and aiding finance teams in reconciliation and invoicing.

ClickStack's journey to general availability involved significant schema optimizations to handle large observability workloads while maintaining performance. The team focused on redesigning primary keys, implementing text indexes, query rewrites, and leveraging ClickHouse features to address slow query issues and optimize for common query patterns. These efforts included benchmarking with ClickCannon to simulate realistic workloads and evaluate resource requirements effectively. Changes such as modifying the primary key strategy, adopting text indexes for better granule pruning, and introducing materialized views improved query efficiencies and reduced latency by more than fivefold. The optimizations also integrated features like alias columns for dynamic attributes, enhancing the overall user experience by making queries more efficient and metadata-driven features more responsive. The result was a robust, scalable schema that not only improved ingestion, storage, and operational efficiency but also aligned with native ClickHouse capabilities, ensuring adaptability to future database enhancements.

ClickHouse Agents, a newly launched fully managed service in ClickHouse Cloud, offers an innovative way to explore and manage Postgres data using natural language queries. This agentic service, powered by Claude and built on the open-source AI chat platform LibreChat, allows users to query both Postgres and ClickHouse simultaneously without writing SQL, enhancing the capabilities for data exploration, query performance analysis, and database monitoring. ClickHouse Agents provide a no-code solution for building custom agents, enabling teams to ask questions in plain English and receive instant answers, thereby facilitating self-serve data exploration and performance tuning. The service supports seamless integration with MCP-compatible systems, allowing users to connect their own agents, models, and tools without vendor lock-in. Additionally, it includes features for monitoring database performance, identifying slow query patterns, and supporting migrations from Postgres to ClickHouse, thus aligning with the vision of a unified data stack for OLTP and OLAP.

Artemis is an AI-native threat detection platform that leverages ClickHouse for real-time analytics across vast amounts of log data to identify potential cyber threats. By implementing innovative solutions such as query coalescing, AI-powered debugging with the Claude Code skill, and materialized extraction, Artemis significantly enhances its detection capabilities and efficiency. These optimizations allow Artemis to process millions of queries rapidly, reducing CPU consumption and improving the speed of investigative queries by up to 60x. The platform's AI agents continuously monitor and analyze data from various sources, providing security teams with insights into anomalous behaviors and potential threats. As the cyber threat landscape evolves, Artemis's ability to adapt and scale without increasing complexity or costs positions it as a formidable tool against AI-powered attackers, with its solutions being trusted by companies like Wix, Mercury, Lemonade, and Upwork.

The ClickHouse 26.6 release introduces several advancements including 56 new features, 79 performance optimizations, and 366 bug fixes, with notable additions such as hypothetical skip indexes, cascading refreshable materialized views, and experimental support for continuous queries. Hypothetical skip indexes allow users to experiment with different index configurations without building them, aiding in performance tuning by providing insights before actual implementation. The cascading refreshable materialized views enhance dependency management by allowing views to refresh based on others without independent timers, reducing latency issues. A new feature enables appending enum values in a simpler fashion, and the CLI now supports inline documentation queries. The update also enhances lightweight and faster query start-up, with improvements in handling deeply nested queries, and introduces streaming queries in an experimental mode, allowing continuous data streaming in queries on Linux platforms. Additionally, clickhouse-local now can listen for external connections, transforming its utility in data analysis contexts.

Jua, a Zurich-based physics AI company, employs ClickHouse Cloud to enhance the speed and efficiency of its physics foundation model, EPT-2, which is used for atmospheric prediction and energy trading applications. By utilizing ClickHouse Cloud, Jua reduced forecast delivery time from an hour to 20 minutes, decreased compute costs by a third, and significantly sped up historical data queries. The company's architecture combines a world model with a continuous learning agent to simulate physical systems from observational data, with the atmosphere being its initial focus. ClickHouse Cloud's operational simplicity and cost efficiency, along with its ability to handle diverse query types at a petabyte scale, made it the ideal choice for Jua, which previously relied on file-based storage systems. This transition has allowed Jua to deliver data faster and more reliably, giving it a competitive edge in a market where timely and accurate data access is crucial. As Jua continues to expand its capabilities into new domains, ClickHouse's offerings, including their observability tools, are set to play a critical role in supporting this growth.

PgBouncer is a single-threaded connection pooler for Postgres that traditionally uses only one CPU core, limiting throughput on multi-core systems. To maximize resource utilization, ClickHouse Managed Postgres employs a fleet of PgBouncer processes proportional to the number of available CPU cores, with each process using the so_reuseport feature to enable kernel-based load balancing of incoming connections across processes. This setup allows clients to connect to a single endpoint without being aware of the multiple PgBouncer processes in use. A challenge with this approach is query cancellation, as a cancel request can be misdirected to the wrong process, but this is mitigated by a peering mechanism that forwards the request to the correct process. In performance testing on AWS EC2 instances, a single PgBouncer process peaked at about 87k transactions per second, utilizing only one core, while a fleet of 16 processes achieved roughly 336k transactions per second by effectively distributing the workload across available cores. The fleet configuration significantly increased CPU utilization and throughput, demonstrating the efficiency of using multiple PgBouncer processes to avoid bottlenecks in high-concurrency environments. ClickHouse Managed Postgres utilizes this configuration by default, optimizing resource use and enhancing performance for scalable applications.

Clever, a leading K-12 education identity platform, transitioned from Datadog to ClickHouse Cloud to enhance its log indexing and analysis capabilities, handling 150 TB of data monthly from over 400 services and 200 AWS Lambdas. This switch allowed Clever to increase log indexing from 10% to 100% and extend log retention from 3 to 60 days without additional costs, achieving a 200x increase in searchable log volume. The infrastructure team developed a custom query layer that translates LogQL to SQL, providing engineers with a user-friendly log exploration experience. Clever's move to ClickHouse involved optimizing schema design for observability, using data skipping indexes, and ensuring multi-region resilience, which collectively improved query performance by 8-10 times. The transformation has enabled efficient log management, facilitating rapid debugging and trend analysis while maintaining cost-effectiveness and high availability across multiple AWS regions.