Metabase VPS Docker Hosting

Metabase is an open-source business intelligence and embedded analytics platform that ships under the AGPLv3 license for free self-hosted use across any team size or industry. The product covers a no-SQL question builder for non-technical users, a native SQL editor for analysts, interactive dashboards with filters and tabs, scheduled subscriptions to email and Slack, alerts on query results or goal lines, and a semantic layer of Models and Metrics for curated datasets across the team.

The official baseline is at least 1 CPU core and 1 GB of RAM for the smallest deployment serving a few users. For every additional 20 concurrent users, Metabase recommends adding 1 CPU core and 2 GB of RAM to the host. A 40-concurrent-user deployment lands at roughly 3 cores and 5 GB of RAM in practice, and 4 to 8 GB of RAM can handle hundreds of users when the JVM heap is tuned through -Xmx settings.

Metabase listens on TCP port 3000 by default for the web UI. Production deployments place an HTTPS-terminating reverse proxy such as NGINX in front of port 3000, with TLS certificates from Let's Encrypt and the upstream pointing at 127.0.0.1:3000 on loopback. Metabase itself does not handle TLS termination natively, so the reverse proxy is the standard production pattern for serving Metabase securely on a public domain.

A small team under 5 concurrent users runs comfortably on 2 GB of RAM, with -Xmx tuned to leave room for the OS and any sibling services on the same VPS. A deployment serving 40 concurrent users wants around 5 GB of RAM and 3 CPU cores per the official sizing guidance. Anything serving hundreds of users moves to 4 to 8 GB of RAM with the heap explicitly tuned through -Xmx in the JVM args.

The Metabase backup model is intentionally simple. Backing up the application database (a PostgreSQL or MySQL dump using pg_dump or mysqldump) is sufficient to restore a Metabase instance from scratch on a clean server. Metabase queries source databases directly without storing any source data on the Metabase side, so the application database holds all the metadata about questions, dashboards, users, and connections that needs preservation in a backup.

Yes. The Metabase Open Source edition is fully free and released under the AGPLv3 license for any self-hosted deployment without seat or capacity fees. The commercial editions (Pro at $575 per month for up to 10 users, Enterprise from roughly $20,000 per year, and the managed Metabase Cloud Starter at $100 per month for up to 5 users) add features such as interactive embedded analytics, row-level security, advanced permissions, and a self-hosted Pro tier.

Metabase stores its own internal application data in an application database that lives separately from your source databases. The default H2 file-based database is for trial and personal use only and will eventually become irrevocably corrupted in any production setting. The official recommendation is PostgreSQL as the application database (preferred) or MySQL/MariaDB as an alternative when Postgres is not an option on the host machine.

Two install paths are official. Run the Metabase JAR directly with a Java 21 runtime, passing JAVA_OPTS or JAVA_TOOL_OPTIONS to cap heap with -Xmx, or run the official Docker image from Docker Hub mapped to port 3000 on the host. In both cases, point Metabase at a separate PostgreSQL or MySQL application database through the MB_DB_* environment variables before the first run completes, since you cannot easily migrate off the default H2 file later without downtime.

Yes, in two distinct roles. Metabase supports PostgreSQL as one of the source databases you query for BI work, alongside MySQL, MariaDB, MongoDB, BigQuery, Snowflake, Redshift, SQL Server, Oracle, ClickHouse, Databricks, Druid, Presto, and many other connectors. PostgreSQL is also the preferred application database for storing Metabase's own internal state such as questions, dashboards, users, permissions, and settings.

Metabot is the AI assistant built into Metabase that generates SQL from natural language inside the SQL editor, fixes broken SQL errors on request, analyzes existing charts in plain language, and builds dashboards from a prompt. It uses a bring-your-own AI key model, with Anthropic supported in 2026 and more providers coming in. Metabase 61 from May 2026 added AI governance with per-group access controls, token and message limits, and customization of Metabot behavior at the team level.