Metabase
Open-source business intelligence — point it at the database your product already writes to and let anyone ask questions of it
Metabase is an open-source business-intelligence tool. You connect it to a database, and people who do not write SQL can ask questions of it by pointing and clicking, while people who do get a SQL editor and dashboards to put the answers on. It has 48,871 GitHub stars as of August 2026 and ships releases most weeks. For a founder the value is specific: it reads the database your own product already writes to, which is where the growth questions that matter actually live.
What Metabase is
Metabase (github.com/metabase/metabase) connects to a database you already have and puts a question-asking layer on top of it. Non-technical people build queries through a visual editor and never see SQL; anyone who wants to write SQL gets a proper editor. Results become saved questions, questions become dashboards with filters, and dashboards can be scheduled out to email, Slack, or a webhook, or have alerts attached so a number that crosses a threshold comes and finds you. It also embeds into your own product, which is how a lot of teams end up shipping customer-facing analytics without building a charting stack.
It is a genuinely old project by open-source standards — the repository dates to 2015 — and it is maintained at a pace most repos never reach: v0.63.14 was published on 20 August 2026, and the commits behind it are ordinary product work rather than housekeeping, including a fix pinning exploration auto-binning to an explicit bin width and a change guarding queue declarations by ownership. The recent work also includes a run of commits trimming an in-product AI assistant, which is a useful signal in itself: the AI layer is real and still being shaped, not a finished feature to plan around.
Why your own database answers the growth questions
Web analytics tells you a visit happened. Product analytics tells you an event fired. Neither knows whether the person did the thing your product exists to do, because that fact is written to your application tables and nowhere else. This is the gap Metabase sits in, and it is why a founder often gets more out of a day with it than a month of dashboard-watching. The questions it can answer are the uncomfortable ones:
- Did they come back? Not "were there returning sessions", but: of the people who signed up in a given week, how many ever did anything on a later day. That is a join across your own tables and it is unanswerable from pageviews.
- Which signups are worth more of the same? Group your real usage by acquisition source and the channel report stops being about traffic. A cohort analysis over product data is a different argument than one over sessions.
- Where does the first session actually stop? The last row a new account wrote before it went quiet is a precise answer to a question that funnel charts approximate.
- What are people asking for that you do not have? If your product stores text your customers wrote, that text is a queryable table, and reading it in bulk is faster than remembering individual conversations.
Getting there is not automatic. Metabase's own documentation is the place to start on connecting a database, and the sane pattern is a read-only user — or a read replica if you have one — rather than pointing a query tool that anyone in the company can use at the primary your product depends on.
The hard part is the question, not the tool
We learned this on our own data at AgentCeres — the AI Growth Officer, at agentceres.com — while trying to answer something that sounds trivial: how many people who signed up ever came back and said something on a later day. The query is short. Getting it to mean what we thought it meant took several attempts, because our own internal test accounts sit in the same tables as customers, and because scheduled prompts our system sends on a customer's behalf are stored in the same shape as messages a human typed. The first version of that query counted both and produced a number that flattered us. Nothing in a BI tool warns you about that — the chart renders just as confidently either way, and a wrong number with a dashboard around it is more persuasive than no number at all.
That is the honest boundary of what a tool like this does for growth. It reads. It tells you, precisely and repeatably, what already happened, and it makes the definition of each metric something written down rather than something re-remembered every time somebody asks. What it does not do is any of the work the answer implies — the follow-up email to the cohort that stalled, the page rewritten for the channel that converts, the outreach to the segment worth more of. That execution layer is what AgentCeres is: a team of AI specialists that drafts and runs the work, with a human approving anything that goes out. Metabase is a good reason to know what to ask them for — a related read is how do I measure if my marketing is working.
FAQ
- Is Metabase free?
- The open-source edition is, and its licence is worth reading rather than assuming. The repository states that source outside the top-level enterprise directory is AGPL, while source inside that directory is under the Metabase Commercial License. The binaries follow the same split: the plain metabase Docker image is the AGPL build, and the enterprise image is the commercial one. There is also a hosted Metabase Cloud plan, which the README presents as the fastest way to start, and you can move between hosted and self-hosted later.
- Do I need to know SQL to use it?
- No, and that is most of the point. The visual query builder covers filtering, grouping, and joining without SQL, which is what lets a non-technical co-founder answer their own questions instead of queuing them behind you. SQL is there when a question outgrows the builder, and in practice a small team ends up with a mix: a few carefully written SQL questions that define the important metrics, and a lot of point-and-click exploration on top of them.
- Should I use Metabase or a product-analytics tool?
- They answer different questions and most teams end up with both. A product-analytics tool is built around events you deliberately instrumented and is better at funnels, retention curves, and session replay out of the box. Metabase is built around the tables your application already writes, which means it can answer questions nobody thought to instrument — at the cost of you writing the definition yourself. If you have to pick one first and your product stores meaningful state, start with the database.
- How much work is it to run?
- Self-hosting is a normal web application with its own database behind it, and the project publishes installation guides for the common paths. The ongoing cost is less about uptime than about curation: dashboards multiply, near-duplicate questions accumulate, and after a few months nobody is certain which of three similarly named charts is the one to trust. Deciding early who owns the definition of each important metric is worth more than any infrastructure choice you make.
You built it. Now grow it.
AgentCeres is a managed AI marketing team — specialists draft the SEO, social, and outreach that fill your links, you approve what ships. 14-day free trial, from $39/month.