Preset vs Apache Superset: When to Self-Host and When to Pay (2026)

Apache Superset is a free, open-source BI tool with a serious SQL editor and 40+ chart types. Preset is the same Superset engine, packaged and hosted by the person who built it. Self-host if you've got DevOps capacity and want to spend nothing on licensing. Pick Preset if you'd rather pay $20 a seat than run Redis and Celery yourself.

That's the one-paragraph version, and it's directionally right. But "DevOps capacity" is doing a lot of work in that sentence. The actual cost of self-hosting Superset, in dollars and in the pager going off at midnight, is bigger than most teams budget for before they try it. Here's the real breakdown.

What's actually different between Preset and Superset?

Nothing, at the code level. Preset runs Apache Superset itself: the open-source project, Apache 2.0 licensed, the same repo your engineers can clone right now. Maxime Beauchemin created Superset at Airbnb (he'd already created Apache Airflow), then founded Preset to run it as a managed product. Log into Preset and you're looking at SQL Lab, the same 40-plus chart types, the same connectors for Postgres, Snowflake, BigQuery, Redshift, and most other warehouses, and the same dataset-and-metrics model people already argue about in GitHub issues.

Preset isn't a fork, and it isn't "Superset-inspired." It's the identical software, minus the part where you provision it yourself.

       Apache Superset (open source, Apache 2.0)
     SQL Lab · 40+ chart types · dashboards · datasets
                         │
         ┌────────────────┴────────────────┐
         │                                  │
   Self-hosted                          Preset
         │                                  │
 Postgres + Redis + Celery         Preset runs the infra,
 workers + TLS + upgrades:         adds SSO/SCIM, RBAC,
 all on you                        alerts, embedding, SLA

What does it take to run Superset yourself, in production?

docker compose up gets you a working Superset in about five minutes. That demo is also where most of the internet's opinion of "how hard is Superset to self-host" comes from, and it's misleading. A demo instance and a production instance are different animals.

The official Docker Compose file wires up six services: a Postgres metadata database, Redis for caching and as the Celery message broker, the Superset web app, an init container that runs migrations and creates the admin user, a Celery worker for async queries and scheduled reports, and a Celery beat scheduler to trigger those reports on time.

services:
  db:
    image: postgres:15
  redis:
    image: redis:7
  superset:
    build: .
    depends_on: [db, redis]
    ports: ['8088:8088']
  superset-init:
    build: .
    command: ['/app/docker/docker-init.sh']
    depends_on: [db, redis]
  superset-worker:
    build: .
    command: ['celery', '--app=superset.tasks.celery_app:app', 'worker']
  superset-worker-beat:
    build: .
    command: ['celery', '--app=superset.tasks.celery_app:app', 'beat']

Six containers isn't the hard part. Most teams already run more than that. The hard part shows up later: HTTPS and a reverse proxy you own, SSO wired through a custom SecurityManager subclass with AUTH_TYPE = AUTH_OAUTH in superset_config.py, Celery worker concurrency tuned so scheduled reports don't queue behind someone's ad-hoc SQL Lab query, and upgrades. Superset's own docs warn about production complexity. Metadata migrations between versions aren't always clean, and skipping a Postgres backup before running superset db upgrade is the kind of mistake you only make once.

None of this is exotic engineering. It's the same operational tax you'd pay running any stateful Python service with a cache layer and a task queue. But the tax is due monthly, not once, and somebody has to own it.

What does Preset add on top of the open-source core?

Preset's free Starter plan already includes more than people expect: unlimited dashboards and charts, all 40-plus visualization types, the SQL editor, and Preset's own semantic layer, capped at one workspace and five users. That's a real product, not a seven-day trial dressed up as one.

Paying changes what you get, not what you're using. Professional, at $20 a user per month billed annually (or $25 month-to-month), removes the user cap and adds a third workspace, role-based access control, scheduled email and Slack alerts, and standard support. Enterprise, priced by quote, is where the infrastructure-avoidance story completes: single sign-on with SCIM provisioning, audit logs, a Managed Private Cloud deployment option, SSH tunnels into on-prem databases, dbt integration, and a support SLA with a dedicated Slack channel. Want to embed dashboards in your own product? That's a $500-a-month add-on for 50 viewer licenses, layered on top of whichever plan you're already paying for.

Every item on that list is infrastructure or process you'd otherwise build yourself: an OAuth integration, a SCIM endpoint, an on-call rotation for a metadata database nobody wants to own.

What stays exactly the same?

The parts people worry about migrating are the parts that don't change. SQL Lab looks and behaves identically. The chart gallery is the same 40-plus types, built on the same Apache ECharts plugin system. Dashboards, datasets, row-level security: none of it changes. If your analysts already know Superset, moving to Preset is a login change, not a training session.

It also makes the decision reversible. Preset doesn't lock your dashboards into a proprietary format, because there isn't one. Outgrow the pricing, or need to move a workload on-prem, and you export the same Superset instance you were already running.

What does each option actually cost?

Self-hosted Superset is free to download and expensive to run properly. Here's a realistic monthly infra budget for a small-to-mid production deployment on AWS or similar:

Call it $200-1,000 a month depending on data volume and how much redundancy you want, and that's still the smaller number. Budget 16-30 hours a month of engineering time for upgrades, worker tuning, and the occasional broken migration. At a $100/hour loaded rate, that's $1,600-$3,000 in labor stacked on top of the infra bill.

Preset skips the infra bill and charges per seat instead:

Run the math for a 10-person data team. Self-hosted lands around $300-500 a month in infra, plus whatever slice of an engineer's time upgrades eat. Preset Professional at 10 seats is roughly $200-250 a month, with zero ops time. Self-hosting wins on pure dollars in two situations: the team is small enough that someone will babysit it for free, or big enough that per-seat pricing stops making sense against a fixed infra bill, somewhere past 40-50 users depending on your cloud costs.

When does self-hosting actually win?

Four situations where I'd tell a team to self-host without much hesitation.

Your infra team already runs Postgres, Redis, and Celery-backed services for other products. The marginal cost of one more workload on infrastructure you're already patching and monitoring is close to zero. You're not starting from scratch, you're adding a tenant.

Budget is the actual constraint, not time. A pre-revenue startup with an engineer who has slack in their schedule and zero dollars for BI tooling should self-host. That's a legitimate trade, not a mistake you'll regret in six months.

You need to modify Superset itself. Custom chart plugins, a non-standard auth backend, white-labeling with zero Preset branding anywhere: that requires the source code, not a subscription.

Data can't leave your infrastructure, full stop. Preset's Enterprise tier offers Managed Private Cloud and SSH tunnels, which cover a lot of compliance requirements. But air-gapped networks and some data-residency rules still leave self-hosting as the only option on the table.

When does Preset win?

Flip each of those and you get Preset's case. No DevOps bandwidth: a five-person analytics team without an infrastructure engineer shouldn't be debugging Celery beat at midnight. Need SSO or SCIM this quarter, not eventually: Enterprise ships it without a project plan attached. Want upgrades to stop being a controlled risk: Preset handles version bumps behind the scenes, and you find out about new features from a changelog instead of a broken migration. And when a team's time is worth more spent on the actual product than on babysitting a BI stack, paying $20 a seat is the cheaper option even when the infra math looks close on paper.

Do you already run Postgres/Redis/Celery-style infra for other services?
├── Yes, and it's cheap to add one more tenant → self-host Superset
└── No, and nobody wants to learn               → Preset
 
Need SSO, SCIM, or an uptime SLA this quarter?
├── Yes → Preset Enterprise
└── No  → either works, run the cost numbers

Preset vs Apache Superset at a glance

For a wider field of options beyond Preset itself (Metabase, Grafana, Lightdash), our Superset alternatives roundup covers the rest of the landscape.

What if you don't want to run Superset at all?

Preset removes the ops burden of Superset. It doesn't remove Superset. You're still working inside SQL Lab, still modeling data as datasets and metrics, still bound by however Superset renders a dashboard. That's the right trade if your team already knows the tool. Skip Preset for this reason alone if you're mainly avoiding self-hosting pain and have never actually used Superset for anything else.

Fastero is a different bet: a SQL editor with schema-aware autocomplete, dashboards an AI agent can build from a plain-English request, and connectors into 20+ databases and SaaS tools it can query across, fully hosted, with nothing to patch and no Celery workers anywhere in the stack. It's not trying to be Superset with a friendlier login page. See how it stacks up against Preset, Metabase, and the rest of the field on our best BI tools comparison, or read how Metabase and Superset compare if Preset's pricing sends you back to the open-source aisle. If self-hosting is the real question, not just Superset specifically, our best self-hosted BI tools guide covers seven other tools worth a look.

FAQ

Is Preset just a rebranded version of Apache Superset? No. Preset runs the actual open-source Superset codebase, same SQL Lab and same charts, hosted and managed by Preset. Maxime Beauchemin, who created Superset (and Apache Airflow), founded Preset to productize it. You're not learning a new tool.

Can I move from Preset back to self-hosted Superset later? Yes, more easily than with most managed products. Because Preset runs unmodified open-source Superset, you can export your dashboards and datasets and stand up your own instance without a rewrite. Migrating the other direction, self-hosted to Preset, is just as clean.

Is Apache Superset really free? The software is, under Apache 2.0. Running it in production isn't: expect $230-560 a month in infrastructure (Postgres, Redis, Celery workers, a load balancer) plus real engineering time for upgrades and incident response. "Free" describes the license, not the total cost of ownership.

What does Preset's free Starter plan actually include? Unlimited dashboards and charts, all 40+ visualization types, the collaborative SQL editor, and Preset's semantic layer, capped at one workspace and five users. It's a genuinely usable free tier, not a crippled trial.

Does Preset support single sign-on? Yes, but it's gated to the Enterprise tier, along with SCIM provisioning, audit logs, and an uptime SLA. If SSO is a hard requirement today, budget for Enterprise pricing rather than Professional.

How hard is it to self-host Superset with SSO already working? Harder than the docs make it sound. You're writing a custom Superset SecurityManager subclass and configuring AUTH_TYPE for OAuth or SAML in superset_config.py yourself, then testing role mapping by hand. It's a solved problem. But it's your team solving it, not a checkbox in a settings page.

Try Fastero free — SQL editor, dashboards, and AI — fully managed, no Docker required. No credit card required.