chmonitor v0.3 — a full rebuild

chmonitor v0.3 is the biggest release since the project started — a ground-up rebuild on TanStack Start. Everything is new: a faster dashboard, an AI agent that answers questions about your cluster, live query monitoring, a data explorer, cluster topology, AI insights, and self-hosting that’s a single command on Cloudflare Workers, Docker or Kubernetes.

Here’s the ~28-second launch film — every scene is the real product:

chmonitor v0.3 — launch film. Dashboard, AI agent, query monitoring, data explorer, topology, health & self-host.

new features

70+

fixes

perf wins

charts

Everything that’s new

Rebuilt on TanStack StartA static-first shell with client-side data fetching — pages load instantly and cache at the edge.
AI agent over MCPAsk your cluster anything. The agent reads system tables through an MCP server and answers in plain language.
Live query monitoringWatch every running and historical query, sort by cost, and drill into the ones that hurt.
Data query explorerBrowse databases, follow the dependency graph, then jump straight into a SQL console.
AI insightsAnomalies and regressions surfaced automatically and ranked by severity.
Metrics & profilerCPU, memory and IO alongside ClickHouse profiler events for real root-cause work.
Query EXPLAIN as a treeThe EXPLAIN plan rendered as an interactive tree instead of a wall of text.
Cluster topologyNodes, shards, replicas and Keeper quorum drawn as a live diagram.

Highlights

A faster dashboard

The whole app moved to TanStack Start with a native Cloudflare Workers bundle. Pages are prerendered as a static shell and hydrate with TanStack Query, so the first paint is instant and data streams in progressively. Multi-host routing stays as simple as ?host=0.

An AI agent that knows your cluster

The new agent connects over an MCP server and can read every system table. Ask “why is this query slow?” or “what changed in the last hour?” and it pulls the metrics, runs the diagnostic SQL, and explains what it found — no more memorising table names.

Health, audit and insights

Color-coded cluster health rolls up into a ready-made audit prompt, while AI Insights continuously scans for anomalies and regressions and ranks them by severity, so the important things float to the top.

Self-host anywhere

A v0.3 deploy is one command. Run it on Cloudflare Workers, Docker or Kubernetes — same codebase, same image, configured entirely through environment variables.

docker compose up -d

Breaking changes

v0.3 is a rebuild, so a few things moved:

URL structure changed from /0/overview to /overview?host=0 (static routing + edge caching).
fetchData() now requires a hostId parameter — it used to be optional.
All data fetching moved client-side; there’s no SSR data layer anymore.

Since the project has no backwards-compatibility guarantees yet, there’s nothing to migrate other than bookmarks.

Changelog

Area	What changed
Dashboard	Rebuilt on TanStack Start; static shell + TanStack Query; 15+ pages, 71 charts
AI agent	New agent over MCP; reads system tables; 29+ tool categories
Query monitoring	Live running + historical queries, cost ranking, EXPLAIN tree
Data explorer	Database browser, dependency graph, SQL console
Insights	AI insights engine — anomalies & regressions ranked by severity
Metrics	CPU / memory / IO + ClickHouse profiler events
Cluster	Topology diagram — nodes, shards, replicas, Keeper quorum
Health	Color-coded health → generated audit prompt
Deploy	One-command self-host on Cloudflare Workers / Docker / Kubernetes
Performance	13 perf wins — pooling, memoization, cache limits, hidden-chart unmounting

See the full commit-level history in the GitHub releases.

Try it now: open the live dashboard, read the docs, or star us on GitHub.