Monitor queries, merges, parts, replication, and health. An AI advisor on top — open source, self-host free.
What you get
Purpose-built views over ClickHouse system tables — queries, replication, storage, and health. No exporters, no extra storage.
AI Agent
The built-in agent reads schema and query_log before recommending. Connect over MCP from Claude, Cursor or any client.
Query monitoring
Running, history, failed, slowest and most expensive queries — duration, memory, rows read and the full statement.
Cluster topology
Interactive map of nodes, shards, replicas and Keeper quorum — live replication links, health states and per-node metrics.
Alerting
Built-in health checks watch replication lag, disk, memory and failed queries — each with tunable warning and critical thresholds.
Data Explorer
Browse every database and table, map dependencies between them, then drop into SQL — all without leaving the dashboard.
Insights
Record-breaking queries, storage stats and a full year of activity — surfaced automatically, no dashboards to build.
PeerDB replication
Point chmonitor at your PeerDB API and every mirror gets live status, rows-synced trends, lag triage and slot health — read-only, proxied, no agent.
Postgres · Beta
Same dashboard, same read-only guarantees. Query insights from pg_stat_statements, live activity from pg_stat_activity, and agent tools — free on every plan while in beta.
Full coverage
From a high-level overview down to ZooKeeper logs and replication mirrors, every system table you reach for, already laid out.
CPU, memory, connections and replication status at a glance, with cluster-wide health checks.
Running, slow, failed and most-expensive queries with full query detail and kill actions.
Record-breaking scans, busiest days and totals, surfaced automatically from query history.
Insert throughput, compression ratios, write amplification and disk write speed over time.
Table sizes, compression and disk headroom, down to part-size distribution.
A live part-lifecycle timeline with merges, mutations and moves by table.
Nodes, shards, replicas and Keeper drawn as a live topology map.
Mirrors and peers with live status, rows-synced trend, CDC batches and slot health.
Recommend-only tuning: skip indexes, projections, PREWHERE and TTL advice, ranked by impact.
Health checks with editable thresholds; Slack, Discord, PagerDuty, Opsgenie, and more; full alert history.
Users, roles, grants and quotas, review who can touch what from one place.
Query, error, trace and ZooKeeper logs plus settings and metrics, searchable and time-ranged.
Why chmonitor
General-purpose tools treat ClickHouse as one data source among many. chmonitor reads its system tables directly and tracks nothing else. Here's how it stacks up against the most common alternatives.
| chmonitor | Grafana + CH plugin | Datadog | CH native /dashboard | |
|---|---|---|---|---|
| Setup time | ~5 min (Docker or CF Workers) | Grafana + datasource plugin install | Agent on each ClickHouse node | Enable in ClickHouse config, restart |
| No exporters / agents needed | reads system tables directly | optional CH datasource plugin | agent required on each node | built into CH |
| ClickHouse-specific depth | Purpose-built: merges, parts, replication, query log, system health | Generic SQL panels, you write the queries | Generic agent metrics; limited CH specifics | Basic built-in panel; minimal detail |
| AI agent + MCP | built-in AI agent; MCP server at /api/mcp | none | AI Advisor on paid plans | none |
| Traffic & ingestion insights | insert performance, compression, write amplification, disk write speed | possible, you build every panel | generic throughput metrics only | none |
| CDC & Postgres monitoring | PeerDB mirrors, snapshot & slot health; Postgres source (beta) | separate datasources, manual dashboards | Postgres integration, no PeerDB | none |
| Alert rules & channels | Built-in health checks, editable thresholds; Slack, Discord, PagerDuty, Opsgenie, and more + per-rule routing and full alert history | Grafana Alerting: many channels and routing rules, once you've built the CH panels | Datadog Monitors: broad integrations, mature escalation policies | none |
| Version-aware queries | SQL adapts to CH schema by version | you manage query compatibility | N/A | fixed to bundled version |
| Cost | Free self-host; cloud in early access | Grafana OSS free; Grafana Cloud paid | ~$15–23 per host per month (varies) | Free (bundled with ClickHouse) |
| Open source / self-host | GPL-3.0; Docker, CF Workers, K8s Helm | Grafana OSS (AGPL-3.0) | SaaS only | within ClickHouse |
Want the deep dive? chmonitor vs Grafana · vs Datadog · vs ClickHouse Cloud
Use cases
Running, slow, failed and expensive queries — with kill and recommend-only EXPLAIN suggestions.
A color-coded health board with Slack/Discord alerts and recovery notifications.
Replica lag, read-only replicas and replication queues, on a live topology map.
Slowest/most-expensive queries plus a recommend-only tuning and capacity advisor.
FAQ
Answers to what evaluators ask most before deploying chmonitor.
docker run -d -p 3000:3000 -e CLICKHOUSE_HOST=... ghcr.io/chmonitor/chmonitor:latest. It also deploys to Cloudflare Workers and Kubernetes via the bundled Helm chart. See the quickstart for step-by-step instructions.SELECT queries against system tables and never writes to your data. The AI agent is also read-only. You control what the connection user can access.More questions? Ask on GitHub Discussions.
Open the hosted dashboard and connect a cluster, or self-host with Docker, Kubernetes, or from source. Self-hosting is always free.