Powering Local AI Agents With Real-Time Analytics

Explore powering local AI agents with real-time databases. Learn how to adapt systems for agent workloads and run a fully local, model-agnostic workflow with Docker and ClickHouse.

Docker ClickHouse LibreChat ClickHouse MCP Server Docker Model Runner

Overview

We’ll look at AI agents as a “new user persona” for real-time databases and explore how systems can adapt to their workloads. Finally, we’ll demonstrate a fully local agentic workflow: a small open model running on-device with Docker Model Runner, connected through the ClickHouse MCP Server to ClickHouse Cloud — zero API keys,

Links

https://github.com/clickhouse/agentic-data-stack
Dockerized LLM stack integrating LibreChat, Langfuse, and ClickHouse MCP.

Tech stack

Docker

Docker is the open-source platform that packages applications and dependencies into standardized, portable containers for consistent execution across any environment.

Docker is the industry-standard containerization platform, enabling developers to build, ship, and run applications efficiently. It uses the Docker Engine (the core runtime) to create lightweight, isolated environments called containers: these units bundle an application’s code, libraries, and configuration. This self-contained approach guarantees consistency, eliminating the 'it works on my machine' problem across development, testing, and production environments (local workstations, cloud, or on-premises). Docker debuted in 2013 and now serves over 20 million developers monthly, simplifying complex workflows like CI/CD and microservices architecture by leveraging tools like Docker Hub for image sharing and Docker Compose for multi-container applications.

https://www.docker.com

View projects
ClickHouse

An open-source, columnar OLAP DBMS: engineered for lightning-fast, real-time analytics on petabytes of data.

ClickHouse is the high-performance, open-source columnar DBMS built for Online Analytical Processing (OLAP). Its architecture, featuring columnar storage and vectorized query execution, is specifically engineered to handle massive data volumes: think billions of rows and petabytes of data. This design enables lightning-fast, real-time analytical queries using standard SQL. It is a proven solution for real-time monitoring, log analytics, time-series data, and large-scale business intelligence (BI) workloads.

https://clickhouse.com

View projects
LibreChat

LibreChat is the enhanced, open-source AI chat platform unifying models from OpenAI, Anthropic, and others in one customizable interface.

LibreChat delivers an open-source, centralized hub for all your AI conversations. It unifies cutting-edge language models from multiple providers (OpenAI, Anthropic, AWS, Azure) in a single, sleek interface . Key features include multimodal support for image and file analysis, a secure Code Interpreter (executing Python, JavaScript, and Go), and advanced agent capabilities . This platform prioritizes privacy and customization, making it the go-to solution for developers and enterprises seeking a secure, self-hosted, and fully extensible AI assistant .

https://librechat.ai

View projects
ClickHouse MCP Server

An integration that connects LLM clients directly to ClickHouse databases using Anthropic's Model Context Protocol.

The ClickHouse MCP Server bridges the gap between AI assistants and real-time analytical databases by implementing Anthropic's Model Context Protocol (MCP). Built on FastMCP, this server exposes structured tools (such as run_select_query, list_databases, and list_tables) that allow clients like Claude Desktop, Cursor, or LibreChat to securely discover, inspect, and query database schemas using natural language. By translating user prompts into optimized read-only SQL queries on the fly, it enables developers and analysts to perform rapid, interactive data exploration directly inside their AI chat interfaces without writing boilerplate integration code.

https://github.com/ClickHouse/mcp-clickhouse

View projects
Docker Model Runner

Docker Model Runner standardizes local LLM deployment by wrapping high-performance inference engines like llama.cpp into portable, OCI-compliant containers.

Docker Model Runner eliminates the 'works on my machine' headache for AI engineering. It packages model weights and inference runtimes into a unified Docker image, allowing developers to spin up local APIs (compatible with OpenAI's schema) using a single command. By leveraging Docker Desktop's GPU passthrough for NVIDIA and Apple Silicon, it delivers near-native performance for models like Llama 3 or Mistral 7B without requiring complex local dependency management. This approach ensures that every member of a dev team runs the exact same model version and environment, accelerating the transition from local prototyping to production-ready microservices.

https://github.com/docker/labs-ai-tools-for-devs

View projects

Powering Local AI Agents With Real-Time Analytics

Related talks