ClickHouse AI

The open platform for AI on real-time data

Bring AI to your live data with managed agents or your own. Get fast answers at agent scale, with an open stack you control.

Get started today View documentation

Trusted by teams at

Why ClickHouse for AI

Put your agents to work on live data.

Run fast at agent scale

Get sub-second answers over billions of rows, even when agents run thousands of concurrent queries. Keep performance and costs predictable as query volume grows.

Use our agents or bring your own

Start with ClickHouse Agents, powered by Claude, or bring your own agents through MCP. Build and run the workflows you need on the same real-time data engine.

Own your stack and trust every answer

Inspect, extend, and run the open components behind your AI stack. Trace and evaluate agent workflows with Langfuse to improve quality, cost, and latency.

Choose your agents.
Keep control of what you build.

Use managed agents, build your own, or connect the ones you already run. Keep the same fast data foundation across every path.

ClickHouse-managed Agents & AI

Assistant · Agent Builder · Code Interpreter · AI Functions

Explore ClickHouse Agents ->->

Bring your own agents

Any agent or framework, over MCP, APIs, CLI & plugins

Connect with MCP ->->

One ClickHouse platform

Sub-second answers at agent scale.

Powered by open projects and protocols you can run

AgentsPublic beta

Chat with live data and ship agents for every team

ClickHouse Agents is the fully managed agentic analytics service in ClickHouse Cloud, powered by Claude. Ask questions in natural language and get answers, charts, and dashboards from live data, then turn your data team’s expertise into governed agents anyone can use.

No-code Agent Builder with skills and context files
Sandboxed code interpreter for Python, JavaScript, and more
SSO, access controls, and encrypted message storage

Explore ClickHouse Agents ->->

ClickHouse Agent

Why did eu-west p95 latency spike on Apr 27? Chart it against the baseline.

Queried ClickHouse· 6,480 rows
Ran Python in the code interpreter· 1.8 s

eu = df[df.region == "eu-west"].p95_ms
base = eu[:"2026-04-27 13:00"].median()
peak, at = eu.max(), eu.idxmax()
back = eu[at:][eu[at:] <= 1.15 * base].index[0]
print(round(peak / base, 1), back - at)

eu-west p95 latency, Apr 27–28 UTC

1,298 ms

baseline 163 ms

Apr 27Apr 28Apr 29

peak vs baseline: 8.0×
to recover: 13 h
peak error rate: 5.9%

AI SQL FunctionsCloud private preview

Put LLMs to work inside your queries

Classify, extract, embed, generate, and translate directly in SQL. AI functions call a language model from any query, so you can enrich, filter, and search text where it already lives, with no pipeline to build and no data to move.

Embeddings and similarity search with aiEmbed and aiSimilarity
OpenAI or Anthropic models, configured once as named collections
Per-query token and API call limits, with retries built in

Explore AI SQL Functions ->->

SQL console

SELECT
  body,
  aiClassify(
    body, ['bug', 'question', 'feature']
  ) AS kind,
  aiExtract(body, 'the product area') AS area
FROM support_tickets
LIMIT 3

support_tickets· 3 rows

body	kind	area
Dashboard crashes when I filter by date	bug	dashboards
How do I rotate my API keys?	question	security
Please add CSV export to reports	feature	reports

MCP

Bring any agent to your ClickHouse data

Connect your agents and tools to live ClickHouse data. Point Claude, Cursor, VS Code, Codex, or your own agent at the ClickHouse Cloud remote MCP server, and it can explore your databases and tables and answer questions with read-only queries.

One remote endpoint, enabled per service
OAuth sign-in, or API keys for headless agents
Open-source MCP server for self-managed ClickHouse

Connect with MCP ->->

Your agentclickhouse-cloud

What's the average session duration by country for the last 7 days?

clickhouse-cloud ·list_tables
clickhouse-cloud ·run_select_query

SELECT country, avg(duration) AS avg_session_s
FROM analytics.sessions
WHERE start_time >= now() - INTERVAL 7 DAY
GROUP BY country
ORDER BY avg_session_s DESC
LIMIT 3

Average session, last 7 days

Netherlands412 s
Germany388 s
Japan351 s

How it works

Your choice of agents. One data foundation.

Use managed AI or bring your own agents to the same ClickHouse data foundation. Build with shared skills and open connections.

ClickHouse welcomes Langfuse: The future of open-source LLM observability

Built for Production AI
Close the AI engineering loop.

Trace full agent workflows, from prompts to tool calls and responses
Evaluate agent outputs with custom scoring and human feedback
Monitor quality, cost, and latency across every LLM call

Get started on langfuse.com

Bring AI to your data

Start a free ClickHouse Cloud trial, or run the open-source components yourself. Use our agents or bring your own, and keep control of your stack.

Start your free trial View the open-source stack