ClickHouse has recently unveiled a suite of open-source tools dubbed "Agent Skills." These aim to steer large language models (LLMs) toward generating more efficient SQL queries, specifically for the ClickHouse database. The initiative, announced in early February 2026, injects AI with 28 prioritized rules covering schema design, query optimization, and data ingestion. The company positions these skills as a means to transform general-purpose LLMs into more adept ClickHouse users, improving performance in production environments.
The core of this development lies in encoding expert knowledge into machine-readable instructions, bridging the gap between LLM-generated code and optimized database operations. This move signals a deeper integration of AI not just for query generation, but for adherence to best practices that can impact database performance and stability. The skills are designed to be compatible with various agentic interfaces, leveraging Anthropic's Agent Skills specification.
Read More: AI in Cybersecurity: New Defenses and New Risks for Businesses
From Generic LLM to Database Specialist
The newly released Agent Skills are presented as a simple add-on for existing AI assistants. Through a single command-line instruction, users can integrate these capabilities, allowing their AI to automatically apply ClickHouse-specific best practices. The system is engineered to detect compatible agent interfaces and deploy the rules accordingly.
Beyond automatic application, these skills can also be manually invoked. The documentation points to examples within environments like Claude Code, where specific commands like /clickhouse-best-practices can be used. This offers a layer of control and direct engagement with the optimization rules.
Open Source and Community Driven
Released under the Apache open-source license, ClickHouse Agent Skills actively encourages community contributions. This open approach suggests a commitment to ongoing refinement and adaptation of the rules based on real-world usage and evolving database needs. The project is hosted on GitHub, providing a central hub for development and collaboration.
Read More: Production Database Slowdown Caused by Small LIMIT Clause in Queries
The skills are designed to assist LLMs in adopting established best practices for working with both open-source ClickHouse and its managed cloud counterpart, ClickHouse Cloud. They are also intended for use with chdb, an in-process version of ClickHouse for Python.
Architecture and Application
The Agent Skills framework appears to be central to emerging ClickHouse architectures, as evidenced by concepts like the "AgentHouse" demonstration and the "Model Context Protocol (MCP)" server. This MCP server acts as a crucial intermediary, translating natural language queries into SQL, managing conversational context, and ensuring secure data access for LLMs interacting with ClickHouse.
The architecture guide for a full-stack AI data platform on ClickHouse highlights how databases are consolidated into ClickHouse, allowing AI agents to query across them holistically. ClickHouse Cloud deployments are also mentioned, with a hosted Remote MCP Server simplifying agent connections without the need for self-hosting the service.
Read More: Build Your Own AI Language Model: Easy Steps For Everyone
Underpinning Principles and Rules
The skills are meticulously crafted, encoding "real-world ClickHouse expertise" into actionable rules. These cover critical areas such as:
Schema Design: Including column ordering in keys, prioritizing filter columns, selecting appropriate data types (native types, minimizing bitwidth,
LowCardinality, avoidingNullable), and effective partitioning strategies.Query Optimization: Focusing on JOIN algorithms, filtering before joining, and leveraging secondary indices.
Data Ingestion: Addressing batch sizing, the use of asynchronous inserts, and alternatives to
ALTER TABLE UPDATEandDELETEoperations.Agent Integration: Establishing connection workflows, schema discovery, and query safety measures like
LIMITand timeouts.
Each rule is accompanied by a description of its impact and application, with specific examples of incorrect and correct practices. The priority order for applying these rules is clearly outlined, emphasizing that ClickHouse's unique architecture necessitates specialized guidance over general database intuition.
Background: The Rise of AI in Data Management
The introduction of ClickHouse Agent Skills comes at a time when LLMs are increasingly being explored for their potential in data management tasks. While LLMs have shown proficiency in generating functional SQL, ensuring that this SQL is performant, efficient, and adheres to best practices in complex database systems has remained a challenge.
This initiative addresses that challenge by embedding domain-specific knowledge directly into the AI's operational framework. By leveraging established patterns and expert-defined rules, ClickHouse aims to democratize the process of optimizing database interactions, allowing a broader range of users to benefit from high-speed analytics powered by AI. The open-source nature of the skills further suggests a collaborative approach to shaping the future of AI-driven data management.
Read More: France Plans Fully Robotic Army by 2040 to Change Future Warfare