Metabase’s superpower is the afternoon. That’s roughly what sits between “we should look at our data” and a working dashboard: point it at a database, and people who’ve never written SQL are asking questions through the visual builder while the analysts drop into the native editor beside them. Nearly 50,000 GitHub stars and 273 million Docker pulls say a lot of teams have had that afternoon. It tops our directory’s Panoply Score, and the reasons are mostly about honesty rather than firepower.
The one-paragraph review
For most small and mid-size companies, Metabase is the correct first BI tool: free if you self-host, cheap and predictable if you don’t, quick to love, easy to leave. Its ceiling is governance: metrics live in questions rather than one modelled layer, and past a certain team size that drift becomes real work. Plan for the ceiling and you’ll still be glad you started here.
Pricing that fits on an index card
The open-source edition is genuinely free: self-host it with unlimited users and you pay only for a server and your own upgrades. Cloud Starter is $100 a month ($90 on annual billing) including five users, then $6 per user; a ten-person team lands around $130 a month, which undercuts a single seat on several competitors. Pro at $575 ($517.50 annual, extra users $12) exists for three specific reasons: white-label embedding, single sign-on, and row-level permissions so customers see only their own rows. If none of those three is your requirement, stay on Starter and bank the difference. Enterprise starts at $20,000 a year for air-gapped deployment and procurement ceremony.
The newer meters are minor but worth knowing: AI features at $3.75 per million tokens, advanced transform runs at a cent or two beyond the included allowance, and result storage at $2 per million rows past the first million. None changes the picture; all are pleasantly written down in public, which is on-brand. Full detail on our pricing breakdown.
Living with it
The visual query builder is the best in class for non-technical users, and the escape hatch to SQL keeps analysts from resenting it. Dashboards, alerts and scheduled reports cover the everyday. Embedding works well even below Pro for internal cases. And operationally it’s a single container: the 273 million Docker pulls back it up: self-hosting this is routine.
Two eyes-open notes. First, the governance ceiling: the same metric built twice will eventually disagree. Metabase knows it, and recent releases have been chipping at exactly this: Transforms (shaping raw tables into curated datasets inside Metabase) arrived in release 59, and release 60 added a metrics explorer, an official MCP server so AI agents can query it properly, and Metabot in Slack, with the AI features open source. It narrows the gap without closing it; the sturdy graduation path is still defining truth in dbt underneath while Metabase stays the friendly face. Second, self-hosting means owning security patches, and it isn’t hypothetical: a widely publicised 0-day in 2023 (which resurfaced on Hacker News this year via the Framework breach disclosure) hit instances that lagged on updates. The Cloud plans exist partly so that’s someone else’s pager.
Running it for the long haul
If you self-host, three habits keep the free lunch free. Update on a schedule: the 2023 0-day mostly hurt instances that lagged, and the fix was shipped fast for the people applying releases. Back up the application database, which holds every question and dashboard your team has built; the container is disposable, that database is the product. And put it behind your identity provider early, because the day someone leaves is the wrong day to inventory who can see revenue.
The graduation pattern deserves a concrete shape too. Start with Metabase straight on the warehouse. When two dashboards disagree about the same metric for the first time, resist fixing it in Metabase: define it in dbt, point the questions at the model, and repeat for every metric that matters. Six months of that and you have governed BI without ever having migrated tools, which is the cheapest governance upgrade in the industry.
Skip it when
Row-level customer isolation, white-label embedding or SSO on day one puts you straight into Pro at $575 a month, and at that price the comparison with Lightdash and friends is worth an afternoon. And organisations that need governed metric definitions before the first dashboard (regulated reporting, hundreds of consumers) are the shape Looker was built for; starting light and graduating later cuts against you when the auditors arrive first.
What the crowd says
G2 sits at 4.4 from 147 reviews, 64% of them five-star and 30% four, with Gartner Peer Insights at 4.2 from a small sample of 15. The gap between its modest review-site presence and its enormous open-source adoption tells you where its users live: engineers who deploy containers don’t write G2 reviews. On Hacker News it’s part of the furniture, to the point that new tools pitch themselves as “a DuckDB-based Metabase alternative” and everyone knows exactly what that means. Being the reference point is its own review.
Verdict
The honest default of the BI category: transparent prices, real free tier, an afternoon to value, and an exit that costs nothing but the dashboards you rebuild. Start here unless you already know you need governed metrics or heavy embedding on day one, and when you outgrow it, do it the graceful way: put dbt underneath rather than tearing Metabase out.
