Metabase 63.19 supports visual questions and SQL
Metabase 63.19 is a web application for querying databases, turning answers into charts, and arranging those charts on dashboards. A visual question builder gives non-SQL users a constrained route into company data, while analysts can use the SQL editor. Alerts, scheduled dashboard delivery, documents, and embedding extend the same saved questions into recurring work.
This breadth comes with a large source tree. Our commit db6e70c checkout held 22,392 files, roughly 1,419,381 lines of source, and occupied 401.9 MB before installation. The backend is Clojure and the browser interface is JavaScript. Both are assembled into one JAR for release, but contributors work across the JVM and frontend toolchains. That is a different commitment from downloading the published JAR and pointing it at a database.
Metabase suits a company where dashboard readers outnumber SQL authors. Permissions can separate internal teams or customers, and database drivers connect the application to operational databases and warehouses. The product also includes transformations, canonical metrics, and content organization. Those features make it more than a chart viewer, which means an owner has to govern database access and decide who may publish trusted definitions.
Self-hosting moves backups, upgrades, and SSL onto your team
The shortest self-hosted route is the official Docker image. The standalone JAR is another documented option. Metabase still needs an application database to store users, permissions, questions, and settings. H2 is available by default, while the developer documentation recommends Postgres. Your analytical data stays in the connected databases, reached through credentials that should have only the access each Metabase group requires.
Our sandbox had 3 CPUs and 8 GB of RAM, but its failed 36-second install tells us nothing about production sizing. We did not start the application, connect a warehouse, or measure dashboard latency. A useful trial needs representative queries, concurrency, permissions, and data volume. Any claim that this machine can support a particular user count would go beyond the supplied run.
Metabase's own cloud comparison is candid about the work. Self-hosters handle upgrades, backups, monitoring, high availability, the load balancer, SMTP, and SSL. The installation guide also says there is no officially supported Helm chart. Cloud removes much of that work, but it does not allow a private code fork or community database drivers. Self-hosting makes sense when regulation, an air gap, a custom build, or a community driver outweighs the operating cost.
What happened when we ran it
Our sandbox cloned commit db6e70c and attempted installation in an unprivileged Node 22 container with 3 CPUs, 8 GB of RAM, and no secrets. The checkout had 96 CI workflow files, a Dockerfile, and a tests directory. Installation failed with exit code 1 after 36 seconds, so the pipeline stopped before build and test steps.
The last lines are unusually specific and still inconclusive. They begin bun install v1.4.2, then run a Node preinstall check. That check prints Please use Bun, not npm or yarn. and exits 1. The package metadata also says Bun is required. The supplied log does not reveal why the script failed to recognize the command that launched it, so assigning blame to Bun 1.4.2, Node 22, or the container would be a guess.
There are no package totals, build timings, passing tests, or failing test counts to report. None of those steps completed. The defensible finding is narrower: this fresh sandbox could not cross the repository's package-manager gate in 36 seconds. A contributor should reproduce the install with the repository's pinned tool setup, then keep any later build and test result separate from this run.
Source work requires Java 25, Node 22 or newer, and Bun
The automated developer setup installs mise and uses it to provision Node.js, Java, Clojure, Bun, and Babashka. The build guide adds uv for Python dependencies and specifies JDK 25. The package manifest requires Node.js 22 or newer and names Bun as its package manager. Windows contributors are told to use WSL, while Apple Silicon builders may need Rosetta 2 for the frontend path.
That stack helps explain why a 401.9 MB checkout is a poor place to improvise versions. bun run dev starts the frontend and backend development flow, and the Clojure CLI fetches backend dependencies separately. Building the release artifact runs both sides and produces an Uberjar. The repository's 96 workflow files show substantial automation, but our run never reached it because local installation stopped first.
Operators can avoid most of the contributor toolchain by using v0.63.19's image or JAR. They still need a production application database, backups, data-source credentials, and an upgrade plan. The release notes explicitly tell operators to back up that application database before upgrading. A published artifact reduces compilation work. It does not turn self-hosting into a stateless container.
An October 5 push and v0.63.19 release show active maintenance
GitHub recorded a repository push on 2026-10-05, and v0.63.19 had shipped four days earlier. The project had 49,541 stars and 4,571 open issues and pull requests. GitHub search counted 3,847 open issues, including current reports about timezone labels disagreeing with tooltips and summarized columns producing empty table cells. The activity is current, and the issue surface is large.
Choose Metabase when colleagues need to answer routine data questions without routing every chart through an analyst. Apache Superset gives analysts a broader exploration environment, Redash keeps authors closer to SQL, and Grafana is a better default for operational telemetry. For Metabase, begin with the release artifact. Our 36-second source failure is a contributor warning, while the lasting operator decision is whether your team wants to own the application database and every upgrade behind it.