Dora NodaGitHub stars are a vanity metric right up until they stop being one. When a single project adds roughly a thousand stars a week for months while its closest rival doubles from a smaller base and a former favorite flatlines, the scoreboard is telling you where a category is consolidating. That is exactly what the September 2026 tallies show for the single-box self-hosted PaaS market, so here is the scoreboard first, before a single paragraph of punditry.
The September 2026 scoreboard
| Project | Sept 2026 stars | 2026 trajectory | Release pace |
|---|---|---|---|
| Coolify | Past 52,000 (tracking ~57,000–60,000 by late September) | ~50,400 in mid-February → ~57,300 by late June, about 1,000 stars a week | Aggressive: stable v4.0 in spring 2026, v4.1.1 by late May |
| Dokploy | Past 32,000 (~36,000) | Roughly doubled from ~19,000 earlier in the year | Rapid pre-1.0 cadence (v0.29.x by May) |
| Dokku | Near 32,000 | A decade-long plateau for the 2013-era CLI veteran | Steady, mature, slow-moving by design |
| CapRover | Stalled around 15,000 | Flat all year | A handful of releases per year; effectively solo-maintained |
Two things make this table signal rather than vanity. First, velocity breaks the apparent 32,000-star tie: Dokploy earned its stars in months, Dokku over more than a decade. Second, star velocity lines up with development pace in every row — the project shipping fastest is gaining fastest, and the project shipping slowest is flat. Stars alone prove nothing; stars plus release cadence tell you where momentum lives.
A note on the numbers: star counts drift by hundreds from week to week, and snapshots taken days apart will disagree in the third significant digit. The ordering above does not depend on any single snapshot — Coolify's first-half trajectory comes from weekly registry snapshots, and every project's rank holds across all 2026 tallies I checked. Treat the exact figures as September readings, and the gaps between them as the finding.
With that established, here is what each row means.
Coolify: the velocity leader
Coolify is no longer just the most-starred self-hosted PaaS; it is pulling away. Weekly snapshots through the first half of 2026 show a nearly straight line: 50,396 stars in mid-February, 51,978 a month later, 52,965, then 55,438 by late May and 57,271 by late June. Late-summer tallies put it between 58,000 and 60,000 with over 5,000 forks and a claimed 481,000-plus self-hosted instances. Nothing else in the category grows like this.
The growth is earned, not hyped. Coolify hit a stable v4.0 in spring 2026 and kept shipping into v4.1.x within weeks. Its library of 280-plus one-click services remains the deepest template catalog of any panel, which matters more than it sounds: most teams pick a PaaS for the ten services they will deploy this year, not the orchestrator underneath. Coolify also became the first self-hosted PaaS to ship an MCP server for AI-driven deployment, a genuinely differentiating move in a year when every platform conversation involves agents. Add multi-server support with experimental Docker Swarm clustering, and Coolify covers everything from a 5-dollar VPS to a small fleet of boxes behind one dashboard.
That breadth is why independent 2026 comparisons keep landing in the same place. Deploynix's own four-way showdown — written by a vendor with every incentive to crown itself — still calls Coolify "the safest all-purpose pick." When even the competition concedes the default, the race for the default is over.
Dokploy: the lean challenger
If Coolify is winning, Dokploy is the reason the win is interesting. Roughly doubling its star count to about 36,000 in 2026 makes it the fastest grower in percentage terms, and it earned that growth with a coherent pitch: the same git-push PaaS experience in a leaner package.
The numbers behind the pitch are concrete. On identical Hetzner hardware, Dokploy idles at about 0.8 GB of RAM against Coolify's 1.2 GB — a 400 MB difference that decides whether a 5-dollar VPS with one or two apps feels roomy or cramped. Its build-source support is arguably broader than Coolify's, spanning Dockerfiles, Compose, Nixpacks, Railpack, and Heroku plus Paketo buildpacks. And its Swarm-native design gives small teams a real clustering story without leaving the panel.
The honest caveat is maturity: Dokploy is still pre-1.0 software (v0.29.x as of May 2026), so part of its velocity is promise-priced. Buyers are betting the 1.0 will land, not observing that it has. Deploynix's verdict captures the equilibrium well: Dokploy is "best if you want a lean Swarm-native panel." That is a real segment, not a consolation prize — but it is a segment, while Coolify owns the default.
CapRover and Dokku: the stall and the steady
The bottom half of the scoreboard tells two very different stories that happen to share a symptom: neither project is gaining stars.
CapRover's stall is the structural one. Star count has sat near 15,000 (tallies range from 14,935 to 15,100) with no visible momentum, and the development pace explains why: a handful of releases per year, effectively a solo-maintained project, a UI that every 2026 comparison describes as dated and occasionally buggy.
The feature gaps have compounded into a checklist — Compose support limited to a custom format instead of full Compose files, no preview deployments, no multi-environment workflow, no managed cloud of any kind. Most worrying is the foundation: CapRover depends on Docker Swarm, which sits in maintenance mode. Swarm works and Mirantis supports it into the next decade, but building a PaaS roadmap on a frozen orchestrator is a bet with a visible ceiling.
Dokku, near 32,000 stars, is the opposite case: stasis as a feature. The 2013-era project invented the self-hosted git-push workflow that every panel on this list now wraps in a GUI, and it still serves the operators who prefer a CLI, Herokuish buildpacks, and zero dashboard. Nobody expects Dokku to ship an MCP server or a template marketplace, and nobody is disappointed that it has not. Holding 32,000 stars without growth is not a stall for Dokku; it is tenure.
The lesson for evaluators: flat stars mean "dying" for a project whose value proposition is feature velocity (CapRover) and "done" for one whose value proposition is stability and simplicity (Dokku). Read the trajectory against the project's own promise, not against the category leader.
What consolidation means, and the ceiling none of them solve
Put the rows together and the market structure is clear: the single-box panel race is being won, not tied. Coolify owns the default pick, Dokploy owns the lean corner, Dokku owns the CLI purists, and CapRover owns steadily less of the conversation. Deploynix's verdict — Coolify safest all-purpose, Dokploy lean Swarm-native, CapRover stability-over-features — reads less like one vendor's opinion and more like the category's settled consensus. The panel-UX and template-library war is converging on a winner the way Heroku-vs-Engine-Yard converged a decade ago.
But here is the thing every star-count comparison elides: all four projects solve "run apps on the box you already have," and none of them solves what happens when one box stops being enough. There is no fleet-wide node provisioning, no declarative machine lifecycle, no story for adding a second server that is more than "install the panel again and squint." Coolify's multi-server support and Swarm clustering stretch the ceiling — you can manage several boxes from one dashboard — but the machines themselves remain pets you provisioned by hand, not cattle described in a manifest and reconciled by a controller.
That ceiling is exactly where the next category boundary forms. A team that outgrows one box does not need a better deploy button; it needs machines that come up from a declared spec, get replaced when they drift, and scale without a runbook full of SSH steps. That is the Cluster API bet: treat the fleet like the workloads, with desired state and reconciliation instead of panels and pets. The star race tells you who won the single-box era. It also tells you, by omission, what the single-box era cannot give you — every project on the scoreboard shares the same ceiling, winner included.
How to choose in late 2026
If you are picking a panel today, the data makes the decision tree short:
- Default to Coolify unless you have a specific reason not to: deepest template library, fastest development, MCP support for agent-driven workflows, and a community large enough that your error message already has a thread.
- Pick Dokploy when footprint is the constraint — a small VPS, a handful of apps, a preference for lean Swarm-native operations — and you are comfortable riding a pre-1.0 release line.
- Pick Dokku when you want the git-push workflow with no dashboard at all and value a decade of stability over new features.
- Think twice before starting anything new on CapRover: it still runs fine for existing installs, but betting a fresh deployment on a stalled, solo-maintained project atop maintenance-mode Swarm is nostalgia, not strategy.
Whichever panel you pick, keep one eye on the ceiling. Stars measure the past year's momentum, not next year's architecture — and the moment your second server arrives, you will be shopping in a different category entirely.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own. Star the repo on GitHub or deploy your first app today.
