Every cluster.
Under control.

One control plane for StarRocks and Apache Doris: cluster operations, query diagnosis, and AI-assisted troubleshooting — shipped as a single static binary.

No agent to install · Embedded database · Six deployment formats

$ curl -LO https://www.xclaw.live/downloads/stellar-1.0.4-linux-amd64-musl.tar.gz

2engines, one console
56diagnostic rules
59REST endpoints
50 MBstatic binary, no runtime deps

From a bare FE address to a diagnosed bottleneck

Stellar reads the cluster the way an operator does: system tables, metrics endpoints and the audit log. Nothing to install on the cluster side.

01Connect

Add a StarRocks or Doris FE address with a read-only monitoring account. Stellar detects the deployment mode and inventories FE, BE and compute nodes.

02Observe

Throughput, latency percentiles, errors, disk and compaction are collected on an interval and kept for trend and capacity analysis.

03Diagnose and act

Query profiles are parsed into an execution graph. Rules name the bottleneck; the AI assistant explains the root cause and proposes a change you approve.

Slow queries, explained in one screen

The profile viewer renders the plan as an interactive graph and marks the operators that dominate the query time. Both StarRocks and Doris profile formats are parsed.

Stellar · Query profile
Query profile graph with AI root cause analysis
Plan graph and root cause. Operator timings on the left, the reasoning chain on the right.
Scan skew IO skew Join explosion Join skew Aggregate skew High-cardinality aggregation Sort spill Exchange bottleneck Fragment imbalance Planner hints Filter efficiency Capacity forecast

56 rules across scan, join, aggregate, sort, exchange, fragment, planner and query families — each with a documented threshold and a suggested action.

Ask the cluster a question

The assistant investigates with real data before it answers, and it can propose actions that require your confirmation.

Stellar · AI assistant
AI operations assistant conversation
Evidence first. Slow queries, disk pressure, load backlog: the assistant gathers evidence, runs the rule engine, then explains — and any risky change needs a human confirmation.

Day-to-day operations, without the SQL console

Cluster inventory, live monitoring, and a SQL workspace that keeps a history of what was actually executed.

Cluster list
Cluster list with health and resource usage
Health, CPU, memory and disk per cluster
Cluster overview
Cluster overview with live metric charts
QPS, latency, errors and resources over time
SQL workspace
SQL workspace with catalog tree and results
Catalog tree, SQL editor, results and history

Built for teams that answer to an auditor

Multi-organization isolation, role-based permissions, and an approval workflow for database accounts — with every change written to an operation log.

Stellar · Permissions
Permission request and approval workflow
Request, approve, then it executes. Account and role changes are requested with a reason, approved by an owner, and recorded.
TENANCY

Organization isolation

Each organization sees only its own clusters and users. Cross-organization visibility is limited to super administrators.

ACCESS

Menu, API and action permissions

Roles are enforced per route and per operation. Permission changes take effect on the next request, without a restart.

AUDIT

Operation log

Who created, changed or deleted what, and when — query history and operation history are kept separately.

Deploy it the way your environment expects

The same statically linked binary ships in every format. State lives in one directory, so an upgrade is a binary swap.

Step-by-step deployment guide →

Static binary

tar.gz

musl build, no runtime dependencies. Extract, run, done.

./stellar server /var/lib/stellar

Debian package

.deb

Creates the service user, installs the systemd unit and enables the service.

apt install ./stellar-server.deb

Container

Docker

Published to GitHub Container Registry. Persist a single volume and you are done.

docker run -v $PWD/data:/data ghcr.io/jlon/stellar

Helm chart

Kubernetes

Probes, resources, ingress and persistence pre-wired for a production install.

helm install stellar deploy/chart

Kubernetes manifests

YAML

Namespace, config map, secret, workload and ingress in a single file to adapt.

kubectl apply -f deploy/k8s/deploy-all.yaml

npm package

.tgz

A distribution channel for machines with Node present. The server itself needs no Node runtime.

npm i -g ./stellar-server.tgz

Questions we get before the first install

What has to be installed besides Stellar?

Nothing. The released binary is statically linked and embeds the web console, the schema migrations and an SQLite database. MySQL and PostgreSQL are supported as external stores by changing one URL.

Can it run in a fully offline network?

Yes. There is no phone-home and no external service required at runtime. The AI assistant is optional and only active if you configure a model endpoint.

Does it need agents on the cluster nodes?

No. Stellar talks to the FE over the MySQL protocol and the HTTP metrics endpoint, and reads the audit table when available. Create a dedicated read-only monitoring account rather than using root.

How do upgrades work?

Replace the binary and restart. Database, uploads and logs live in one data directory; schema migrations run automatically at start-up.

What is the license?

Apache-2.0. Run it internally, modify it, redistribute it — including commercially.

What's new in v1.0.4

Released 2026-10-30 · downloads served directly from this site

01Offline activation

Air-gapped deployments can now be activated with a code signed against the machine fingerprint and verified locally. Migrations ship for SQLite, MySQL and PostgreSQL.

02Cluster diagnostics

Node diagnostics render memory and cache metrics as themed charts. Cluster HTTP clients no longer inherit the ambient egress proxy, which surfaced as "upstream unavailable" on private addresses.

03Console polish

Draggable sidebar ordering, accessibility and contrast fixes across 23 pages, and the About panel now reports the real build version instead of a hard-coded one.

Your first cluster, up in a few minutes

Download a single file, start it, open the console. No dependencies to resolve and no package manager required.