Apache dashboards: what a small team should actually run

Superset can give a small business useful shared dashboards, Zeppelin suits an analyst who writes code, and Druid needs a clear reason for fast event analysis; all require IT help. The other databases in this group are usually a provider's responsibility. The chart libraries belong inside software your developer builds.

Begin with the decision a report will change. A bakery comparing waste between two shops needs agreed definitions of sales and waste before it needs a new database. A delivery company investigating late scans may need much faster event queries. Those are different purchases, even if both end with a chart.

We checked the deep topic releases and official documentation on October 29, 2026. No product was installed for this article. Every business below is an illustration. Incubating entries are projects still undergoing Apache incubation and are not yet fully endorsed by the ASF; that label is not a production readiness certificate.

October 30, 2026 update: added setup time and costs when an AI agent does the typing.

Every project in this group

Registry labels refer to the ASF project directory data, read October 29, 2026 at 21:00 UTC. Component, preview and file date qualifications are part of the release description.

Superset: a shared answer to an ordinary business question

Superset connects to databases and lets people build charts, dashboards and SQL queries. It does not replace the database holding your orders. Someone has to give it a useful dataset, an account with appropriate permissions and definitions that staff understand. The hardest part of a first dashboard is often deciding whether a canceled order still counts as a sale.

For example, a bakery with two locations and fourteen staff could describe its problem as baking the wrong amount every Monday. A provider could connect daily item sales and recorded waste to a read only reporting dataset. The manager would review a short comparison before placing the flour order. Our five user low case below costs $457 a month plus $1,634 setup for self hosting, compared with $125 a month for Preset Professional. Data cleanup and dashboard design are separate on either route.

For example, a nine person HVAC contractor could want to see why completed jobs remain uninvoiced. Its provider could publish job completion date, invoice date and responsible office role, keeping customer contact details out of the dashboard. Each morning the office manager would work the oldest exceptions. The ten user high case is $1,141 a month plus $4,085 setup, versus $250 a month for Preset. The extra time in the high case is our allowance for operating work, not a claim that ten users inherently consume that much administration.

The demo container is not the production plan

The Superset FAQ reports community experience of 8 GB RAM and two virtual CPUs for a moderately sized instance. That is useful sizing guidance, not a universal minimum. We use one 4 vCPU, 8 GiB server as our assumption for a small pilot with an existing source database. Both low and high cases have one host. The source database's existing costs are excluded equally from both alternatives.

The Docker Compose guide explicitly says its examples are not recommended for production. It points single host users toward minikube and the Kubernetes installation guidance. The current moving documentation points to an operator and marks the older Helm chart deprecated. Ask your provider to match the deployment instructions to the release being installed. Container startup, authentication and a recoverable metadata database are separate jobs.

The distribution directory lists 6.1.0. Its May 13, 2026 date is a distribution file timestamp, not an independently confirmed release announcement. This article preserves that distinction. The provider needs Python application and database skills, plus the ability to test access by role. A read only database account is a useful starting restriction, but a dashboard can still disclose sensitive information to the wrong viewer.

Managed Superset wins this small team comparison

Preset hosts Superset. Its Starter plan is free for up to five users; Professional is $25 per user per month billed monthly, or $20 paid annually. We use monthly Professional because the illustrated teams want its additional controls and scheduled reports. Check whether Starter is enough before paying. Our assumption is five users low and ten high, zero separately charged managed setup for this subscription comparison, and no embedded dashboard viewer add-on.

Setup = hours × $68.09. Monthly = servers + (stored GB minus 10) × $0.00695 + administration hours × $68.09. Total = setup + 36 × monthly; savings = alternative total minus self hosted total. Calculations use unrounded values; displayed totals round to whole dollars.

The hosted subscription costs less in both cases. An outside provider's bill will normally exceed the employee compensation rate used here, which makes the case for self hosting weaker unless you have a particular access, hosting or integration requirement. Get a quote. If staff already have a suitable report service, use that before creating another place to manage accounts.

Start with the official installation guide and one dashboard from invented or non-sensitive sample rows. Test it as a viewer who should see only one location. Export its definitions, restore the metadata database into a fresh environment and confirm that it still answers the same question. Skip Superset when your existing point of sale report already provides the answer or nobody can maintain the query behind the chart.

Setup time when Claude or ChatGPT does the typing

We keep the Superset example's one $48 server and 100 to 500 GB of backups. We apply our IT assistance assumptions to its existing IT labor budget, despite using one host. These paired low and high cases are assumptions, not measured installations; human review is included. Coding studies do not validate deployment savings.

With L = $68.09/hour, S/A = manual setup/monthly hours, q/r = AI shares, H = hosting plus backups, and T0/Tm = setup/monthly tools: manual total = S × L + 36 × (H + A × L); AI total = S × q × L + T0 + 36 × (H + A × r × L + Tm). Round totals after calculation.

Claude and ChatGPT each cost $100 for the selected monthly plan: one setup month, then $50/month (50%) for 36 operating months if other work uses the rest. Dedicated use adds $1,800 to each total. Allowances differ; check limits and budget extra usage. High cases save no admin hours; subscriptions exceed manual costs from month 7.

API billing is separate. Token cost = (ordinary input × rate + cache writes × rate + cache reads × rate + output × rate) / 1,000,000. Our cumulative budgets include retries and billed reasoning: low input is 30% ordinary, 10% short writes, 60% reads; high input is uncached. Use standard rates within context limits.

Claude Code or Codex can draft Superset deployment settings and metadata backup scripts. Desktop Commander can install dependencies, edit Superset configuration and read worker logs through authorized server access; we budget its free local server, excluding optional paid Remote. Claude in Chrome can inspect provider settings, Cloudflare DNS and Superset viewer permissions; ChatGPT Work offers a separate cloud browser, not an assumed Chrome extension.

Use a new or staging server, keep secrets out of chat, and approve commands or restrict them in a sandbox. Watch dashboard actions; review changes, test viewer permissions and restore the metadata database. Use subscriptions through supported supervised tools and API keys for custom automation; check applicable terms and never share account credentials. We still own backups, updates and access decisions.

Our conclusion stays the same: Preset Professional remains cheaper at $4,500 to $9,000 over 36 months; AI assistance does not pay for dashboard design.

Druid: investigate events while the answer still matters

Druid's introduction describes an analytics database designed for fast aggregation of event data. Examples include activity records and time based measurements. It is a poor starting point for ordinary monthly reporting when a conventional database query finishes quickly enough. You also need to budget for bringing the events into it and retaining the source data needed to rebuild them.

For example, a thirty person delivery company could need to find which depot has stopped producing parcel scans during the current shift. Its data provider could evaluate Druid on synthetic scan events and a query grouped by depot and time. The dispatcher would see a growing gap before the end of the day. The low self hosted evaluation allowance below is $457 a month plus $1,634 setup. That buys an operating budget for a private trial, not a production dispatch system or a promise about query speed.

For example, a forty person ecommerce operation could investigate repeated checkout failures during a promotion. Its provider could load timestamped error categories without card data and test how quickly an analyst can isolate the failing step. The normal daily change would be a shorter route from an error alert to an explanation. The high trial allowance is $1,141 a month plus $4,085 setup. Check PCI DSS before allowing payment data into logs; this example deliberately uses only nonpayment event categories.

The current local quick start needs six GiB

The local tutorial specifies at least 6 GiB RAM, a Unix like system, Java 17, Python 3 and Perl 5. The single server guide now favors the automatic startup script. Its old four GiB nano example is explicitly a deprecated, limited evaluation configuration. Do not quote that older example as the current tutorial minimum.

Our low and high cases both use one 4 vCPU, 8 GiB host and 100 to 500 GB of retained backups. This is an evaluation sizing assumption that clears the tutorial's memory floor. It does not size a resilient production service. The IT provider needs JVM operations, ingestion design, storage and recovery skills. Moving to production means agreeing availability, replicas, durable storage and authentication before repricing the system.

The download page identifies 38.0.0, released October 1, 2026. Start there and use the matching tutorial. Run an evaluation with a bounded copy of events and compare counts back to the source. A fast query against an incomplete import is still a wrong report.

A useful floor, not a finished managed quote

Imply Polaris pricing starts at $100 a month for Starter, described as a proof of concept environment limited to 25 GB with variable performance. Standard production environments start at $600 a month. Billing depends on project usage, ingestion and storage. Neither starting figure is a complete quote for the businesses above.

Setup = hours × $68.09. Monthly = servers + (stored GB minus 10) × $0.00695 + administration hours × $68.09. Total = setup + 36 × monthly; savings = alternative total minus self hosted total. Calculations use unrounded values; displayed totals round to whole dollars.

This table prices only the self hosted evaluation. The vendor FAQ requires an upgrade to Standard or Custom after the trial, so a three year paid Starter scenario is not verified. Standard's advertised starting floor would be 36 × $600 = $21,600 before usage and any setup. It is not comparable to a nonresilient trial server. Actual managed spending, savings and break even therefore require a production design and a vendor quote. Do not turn a short evaluation price into a long term production saving.

Skip Druid when daily exports are timely enough or your team has no one to own ingestion failures. Ask the provider to test late arrivals, duplicate events and a rebuild from retained source data. If a missing hour changes the conclusion, the dashboard should say that data is incomplete.

Zeppelin: give an analyst a reproducible notebook

Zeppelin combines code, queries, results and charts in notebooks. It fits someone who explores a question and wants to keep the steps alongside the answer. It also allows code execution, which is a reason to restrict access carefully. A notebook service is more powerful than a read only dashboard and needs a different permission review.

For example, a seven person print shop could want to compare estimated material use with actual consumption on repeat jobs. A part time analyst could use a notebook to read exported job records and explain the calculation in adjacent text. The owner would see which jobs deserve a revised estimate. Our five viewer low cost case is $117 a month plus $272 setup on one server; Power BI Pro for five users is $70 a month paid yearly. The notebook has value only if the analysis needs code that the report service does not cover.

For example, a ten person farm and winery operation could compare packing loss between harvest batches. Its analyst could keep the import, unit conversions and chart in one notebook using non-sensitive production records. At the weekly planning meeting, staff could trace a surprising total back to the original batch. The ten user high allowance is $256 a month plus $817 setup, compared with $140 a month for Power BI Pro. Neither price includes designing the analysis or cleaning the source records.

Keep the notebook and its dependencies together

The install guide lists JDK 11 and its tested operating systems. It publishes no universal RAM or CPU minimum. Additional interpreters can require their own software, such as Spark or Flink. Our single 4 vCPU, 8 GiB machine is a sizing assumption for a limited SQL and small file pilot, with no separate compute cluster. Low and high host counts are both one.

The download page confirms 0.12.1 dated June 12, 2026 and supplies an official container example for that version. The moving install guide still shows older example image tags. Follow the current release page rather than copying an old tag without checking it. Persist notebooks and configuration outside the disposable container, and make sure the restore instructions include interpreter dependencies.

The analyst needs SQL or programming skills. The operator needs Linux or container administration and must decide which interpreters users can run. Our 4 to 12 setup hours and 1 to 3 monthly admin hours apply only to this limited single server pilot. Broader access, additional clusters and regulated data require a different budget.

Setup = hours × $68.09. Monthly = servers + (stored GB minus 10) × $0.00695 + administration hours × $68.09. Total = setup + 36 × monthly; savings = alternative total minus self hosted total. Calculations use unrounded values; displayed totals round to whole dollars.

Power BI Pro is $14 per user per month paid yearly, confirmed October 29, 2026. The table assumes five to ten paid users and zero separately charged subscription setup. Power BI is a reporting alternative, not a hosted Zeppelin notebook; choose it when the needed output is an ordinary shared report. Its subscription is cheaper in both cases.

Start with the official release and installation links. Save an input file, run the notebook, restart the service and reproduce the result. Ask another person to follow the notebook without the author's help. Skip Zeppelin if nobody is comfortable reading the calculation or if people only need to view a fixed chart.

Needs your IT department or a provider

Cloudberry (Incubating). Its platform requirements recommend at least two production coordinators with 16 or more cores and 32 GB each, plus two segments with eight or more cores and 32 GB each. Single node deployment is for development and testing. A forty five person wholesaler with an existing data team might justify it for years of order history. Assume 6 to 16 monthly operations hours, plus data engineering. A small business should price BigQuery first. The shared eight vCPU tier cannot meet the coordinator recommendation, so we do not invent a cluster total.

Doris. The 4.x environment guide permits a small test layout but recommends separate production frontend and backend services, including three backends; its production hardware recommendation is 16 or more cores and 64 GB per machine. A thirty person delivery firm might use it for operational reports. Assume 6 to 16 administration hours/month plus ingestion work. VeloDB Cloud is a direct managed alternative: Standard US East compute is $0.134/vCPU/hour, storage $0.00005/GB/hour and cache $0.000332/GB/hour, read October 29. Those are usage units, not a complete monthly bill. Get a workload quote rather than pricing undersized machines.

Kylin. The prerequisites recommend a 16 vCore, 64 GB, 500 GB Kylin server alongside Hadoop services. Its single node mode still relies on surrounding compute, storage and a metadata database. A twenty five person manufacturer with an existing warehouse could justify it for repeated margin reports. Assume 6 to 16 operational hours/month for an already small platform, excluding report development and on call service. BigQuery is an alternative warehouse approach, not a hosted Kylin instance. Starting a Hadoop estate merely for a monthly margin report is unlikely to make sense.

Pinot. Its quick start is a local or Docker trial. Deployment guidance describes controller, broker and server roles, a recommended three node ZooKeeper ensemble for production and durable deep storage. No universal CPU/RAM minimum applies. A forty person ecommerce team building customer facing event dashboards could justify a provider. Assume 6 to 16 monthly operational hours plus integration. StarTree is the managed candidate: its price page lists $0.21 per production vCPU hour fully managed in its cloud, with no separate ingest or storage charges, or $0.11 per production vCPU hour run inside your own cloud account plus that provider's infrastructure bill (list prices, read October 29, 2026). BigQuery may fit a less time sensitive reporting requirement, but no latency equivalence is claimed.

BigQuery's US Iowa on demand query price is $6.25 per TiB after the first 1 TiB per month free, read October 29, 2026. Storage is additional. For Cloudberry, Kylin or a simpler Pinot reporting need, that unit price provides a starting comparison; query volume and storage must be filled in before claiming monthly savings. No one of these projects removes the need for an analyst who understands the source records.

Your developer's toolbox

The four developer entries above appear inside applications or define formats used by reporting tools. ECharts draws interactive charts in a site or app. DataSketches estimates aggregate statistics with bounded data structures; ask whether a displayed count is approximate before using it for reconciliation. Flagon supports study of how people use software, which makes collection scope and privacy review part of the design. Ossie (Incubating) defines a shared format for business metrics so reporting tools can interpret them consistently. Its catalog has no verified release, so require a concrete supported version before depending on it. None is a ready made owner dashboard service.

Risks, support and ownership

Read the ASF project security directory for each project's advisory route and the ASF security guidance. Read the project guidance for Superset security, Druid advisories and Zeppelin security. Keep credentials outside notebooks, limit database permissions and test exports as an ordinary user. A picture can disclose customer information just as a spreadsheet can. Check HIPAA before using patient information, attorney client confidentiality before legal matter data, and FERPA before education records; no configuration described here establishes compliance.

Check the last release, recent mailing list activity and the Attic before adopting a quiet project. Preset, Imply and VeloDB describe their services on the vendor pages cited above; ask for support hours and restore responsibilities in writing. Community discussion is useful but does not promise incident response.

The Apache License 2.0 permits business use and modification subject to its terms, including notice obligations when distributing covered work. It does not supply a warranty or a broad trademark license. This is an operational explanation, not legal advice.

Download the kit

Download the analytics kit for provider cards, first report checks and editable cost worksheets.

Related reading

Use the Apache project guide to choose a group, then read workflow automation, databases and big data platforms. The n8n guide can help when reliable exports are the immediate missing step.

All amounts are US dollars before tax, read October 29, 2026. Shared inputs: DigitalOcean $48/month for 4 vCPU and 8 GiB; Backblaze B2 $6.95/TB/month after 10 GB free; BLS administrator wage $47.66/hour divided by the compensation share 0.700 gives $68.09/hour rounded. Backup assumptions are 100 to 500 retained GB with no chargeable transfer/API usage. All hours are our low and high assumptions. Custom development, migration, training, compliance and incidents are excluded; a provider charges above this employee compensation proxy.

Fact-check notes and sources

Official pages below were read October 29, 2026. Catalog rows and source entries without an individual time use the shared directory research from 21:00 to 21:27 UTC; additional IT research ran from 21:37 to 21:42 UTC. Later checks are recorded individually. Registry dates can lag; file dates identify distribution timestamps, not release announcements. We read documentation and pricing; we did not install these products.

1. Incubation Policy, read October 29, 2026, 21:37 UTC.
2. Apache Cloudberry (Incubating), read 2026-10-29, shared series research.
3. DataSketches |, read 2026-10-29, shared series research.
4. Apache Doris: Open Source, Real-Time Analytics and Search Database for the AI Era, read 2026-10-29, shared series research.
5. Apache Druid, Apache® Druid, read 2026-10-29, shared series research.
6. Apache ECharts, read 2026-10-29, shared series research.
7. Apache Flagon, read 2026-10-29, shared series research.
8. Welcome to Apache Kylin, Apache Kylin, read 2026-10-29, shared series research.
9. Home, Apache Ossie (incubating), read 2026-10-29, shared series research.
10. Apache Pinot, Real-Time Analytics for Users and AI Agents, read 2026-10-29, shared series research.
11. Welcome, Superset, read 2026-10-29, shared series research.
12. Zeppelin, read 2026-10-29, shared series research.
13. projects.apache.org, projects.json, read 2026-10-29T21:00:00Z, shared directory snapshot.
14. superset.apache.org, intro, read October 29, 2026, 22:10 UTC.
15. superset.apache.org, faq, read October 29, 2026, 21:39 UTC.
16. superset.apache.org, docker compose, read October 29, 2026, 21:39 UTC.
17. superset.apache.org, kubernetes, read October 29, 2026, 21:39 UTC.
18. Index of /superset, read October 29, 2026, 21:38 UTC.
19. Pricing, Preset, read October 29, 2026, 21:38 UTC.
20. druid.apache.org, quickstart, read October 29, 2026, 21:47 UTC.
21. Local quickstart, Apache® Druid, read October 29, 2026, 21:47 UTC.
22. Single server deployment, Apache® Druid, read October 29, 2026, 21:47 UTC.
23. Download, Apache® Druid, read October 29, 2026, 21:38 UTC.
24. Imply Database as a Service Pricing, read October 29, 2026, 21:47 UTC.
25. Apache Zeppelin 0.12.0 Documentation: Install, read October 29, 2026, 21:38 UTC.
26. Download, read October 29, 2026, 21:38 UTC.
27. Power BI: Pricing Plan, Microsoft Power Platform, read October 29, 2026, 21:39 UTC.
28. Platform Requirements, Apache Cloudberry (Incubating), read 2026-10-29, shared series research.
29. Apache Cloudberry Deployment, Apache Cloudberry (Incubating), read October 29, 2026, 21:37 UTC.
30. Hardware and Software Environment Check, Apache Doris, read October 29, 2026, 21:39 UTC.
31. Pricing, VeloDB, read October 29, 2026, 21:39 UTC.
32. Prerequisite, Apache Kylin, read October 29, 2026, 21:39 UTC.
33. Single Node Mode, Apache Kylin, read October 29, 2026, 21:39 UTC.
34. Quick Start Examples, Apache Pinot Docs, read October 29, 2026, 22:10 UTC.
35. Deployment, Apache Pinot Docs, read October 29, 2026, 21:42 UTC.
36. BigQuery, Google Cloud, read October 29, 2026, 21:39 UTC.
37. Apache Project Security Information, read October 29, 2026, 21:47 UTC.
38. ASF Security Team, Apache Software Foundation, read October 29, 2026, 21:37 UTC.
39. superset.apache.org, security, read October 29, 2026, 22:06 UTC.
40. Apache Druid security advisories, read October 29, 2026, 22:07 UTC.
41. Security, read October 29, 2026, 21:58 UTC.
42. The Apache Attic, The Apache Attic was created in November 2008 to provide process and, read October 29, 2026, 21:47 UTC.
43. apache.org, LICENSE 2.0, read October 29, 2026, 21:37 UTC.
44. Droplet Pricing, DigitalOcean, read 2026-10-29, shared series research.
45. B2 Cloud Storage Pricing, Backblaze, read 2026-10-29, shared series research.
46. Occupational Employment and Wage Statistics Profiles, read 2026-10-29, shared series research.
47. Bureau of Labor Statistics Data, read 2026-10-29, shared series research.
48. Claude plan pricing, read 2026-10-29T23:31:08.854Z; supplied AI setup research.
49. Claude Max plan, read 2026-10-29T23:31:40.711Z; supplied AI setup research.
50. Desktop Commander pricing, read 2026-10-29T23:31:08.785Z; supplied AI setup research.
51. Desktop Commander setup and permissions, read 2026-10-29T23:31:41.305Z; supplied AI setup research.
52. Claude in Chrome setup, read 2026-10-29T23:31:11.010Z; supplied AI setup research.
53. Claude in Chrome permissions, read 2026-10-29T23:31:09.885Z; supplied AI setup research.
54. Anthropic Consumer Terms, read 2026-10-29T23:31:08.765Z; supplied AI setup research.
55. METR study of experienced developers, read 2026-10-29T23:30:58.118Z; supplied AI setup research.
56. OpenAI prompt caching, read 2026-10-29T23:32:00.479Z; supplied AI setup research.
57. OpenAI API pricing, read 2026-10-29T23:31:13.960Z; supplied AI setup research.
58. Codex CLI, read 2026-10-29T23:32:01.197Z; supplied AI setup research.
59. Codex with a ChatGPT plan, read 2026-10-29T23:31:16.001Z; supplied AI setup research.
60. ChatGPT Work cloud browser, read 2026-10-29T23:32:01.431Z; supplied AI setup research.
61. Anthropic API pricing, read 2026-10-29T23:31:14.425Z; supplied AI setup research.
62. ChatGPT plan pricing, read 2026-10-29 about 23:27 UTC; supplied AI setup research.
63. OpenAI Terms of Use, read 2026-10-29 about 23:27 UTC; supplied AI setup research.

Apache, the Apache feather logo and the Apache project names are trademarks of The Apache Software Foundation. This site is not affiliated with or endorsed by the ASF. Other product names are used for identification only.