Dora NodaEvery 2026 roundup of self-hosted Heroku alternatives opens with the same two numbers: Coolify's GitHub star count and the size of its one-click template catalog. The implication is that the contest is already decided — the most-starred project with the most templates wins, and the only remaining question is how far behind the runners-up are. That framing is wrong in an interesting way. Stars and template counts measure something real, but what they measure is day-one developer experience, not what the platform does for you on day two hundred. And on day two hundred, all three leading projects hit the same ceiling.
The Scoreboard, All on One Page
Here are the June 2026 numbers, verified against each project's repo and release pages rather than quoted from memory:
| Coolify | Dokploy | CapRover | |
|---|---|---|---|
| GitHub stars (June 22, 2026) | ~57.3K | ~35K | ~15.1K |
| GitHub stars (Aug 13 re-check) | ~60.5K | ~36.6K | ~15.1K |
| One-click catalog | 280+ services | 500+ blueprints | 350+ one-click apps |
| Latest stable (mid-2026) | v4.0.0 (Apr 27, 2026) | v0.29.x (pre-1.0) | v1.14.2 (May 2026 hotfix) |
| License | Apache 2.0 | Apache 2.0 core + source-available paid features | Apache 2.0 |
| Managed control plane | Coolify Cloud, $5/mo base (2 servers, +$3 per extra) | Dokploy Cloud from ~$4.50/server | None (self-hosted only) |
| Under the hood | Docker on servers you supply, orchestrated from one panel | Single-node Docker Swarm + Traefik | Docker with long-standing Swarm support |
| Project age | Since 2021 | Since 2024 | Since 2017 |
Two things in that table should surprise anyone who only read the headlines. First, the ordering never moves: an August re-verification against the GitHub API showed the same 1-2-3 ranking, with Coolify still adding stars fastest. Second, Coolify does not actually win on catalog size — Dokploy's blueprint library and CapRover's one-click collection both outnumber its 280+ services. Catalog size stopped being a moat some time ago; nobody's marketing page has caught up.
Sources for the figures above: the Cloudzy Coolify-vs-Dokploy comparison (stable versions, star counts, license split), the Deploynix 2026 showdown (template counts, managed-plane pricing, CapRover's v1.14.2 security hotfix), and a Stormkit roundup PR that re-verified every star count and license against the GitHub API on August 13, 2026.
What 57K Stars Actually Measure
Stars measure how many developers had a good first hour, compounded over time. Coolify earns them honestly: a one-command install script, a polished web UI, git-push deploys with HTTPS by default, and weekly releases from a solo-founder-led project with thousands of paying cloud customers behind it.
Dokploy's steeper velocity curve — from zero to ~35K stars in about two years — measures the same thing for a slightly different audience: developers who prefer its cleaner UI and native Compose feel. CapRover's flat ~15K measures age and stability rather than momentum; it has been boring and reliable since 2017, and its May 2026 security hotfix proves it is still maintained, just no longer accelerating.
None of that measures architecture. A star is cast at the moment the demo works on a single VPS — before the voter has run a second server, survived a node failure, or tried to drive the platform from an AI agent instead of a browser tab. Star velocity answers "which onboarding is smoothest," which is a real question. But it is not the question "which platform survives my team growing past one box." Conflating the two is how teams end up re-platforming under pressure eighteen months later.
Release maturity tells a complementary story the star counts blur. Coolify's v4.0.0 (April 2026) is a declared stable release; Dokploy is still pre-1.0 in the v0.29.x line, which means its fast iteration comes with fast-breaking-change risk; CapRover ships rarely but each release lands on nearly a decade of production surface. If your selection criterion is "least likely to surprise me," the ranking inverts: the veteran first, the stable v4 second, the pre-1.0 challenger last.
What 280+ Templates Hide
A one-click template turns "deploy Postgres" into a button. That is genuinely valuable — it compresses thirty minutes of Compose-file writing and secret-wiring into thirty seconds, and for a solo developer standing up side projects it may be the single most-used feature in any of these three dashboards.
But every template in every one of the three catalogs deploys the same shape: containers on one host (or one Swarm), with local volumes, behind Traefik or Nginx, managed from a panel. The catalogs compete on breadth of day-one choice, not on depth of day-two lifecycle.
Five hundred blueprints for single-host Postgres do not add up to one story for Postgres that survives losing its host. Three hundred one-click apps do not include one-click failover, one-click cross-node storage migration, or one-click tenant isolation — because those are properties of the orchestration layer underneath, and all three tools share effectively the same one: Docker, optionally Swarm, on servers you provisioned by hand.
This is the precise sense in which template count hides the gap. It makes three architecturally similar platforms look maximally differentiated on the axis that matters least at scale (how many apps can I install in one click) while leaving the axis that matters most at scale (what happens when one box is not enough) entirely off the comparison. A team choosing on template count is optimizing the first weekend and ignoring the second year.
Three Gaps No Popularity Metric Covers
Multi-node scheduling. Coolify orchestrates independent Docker hosts from one panel and labels Swarm mode experimental in its own docs; Dokploy defaults to single-node Swarm so workers can be added later, but the scheduling model underneath is still Swarm's; CapRover has the longest Swarm history and the slowest-moving UI on top of it.
Swarm restarts a failed service on a surviving node — a reactive restart, not a reconciled recovery with health-check-driven rollback. None of the three gives you namespace-scoped tenants, topology-aware placement, or storage that follows the workload. Kubernetes-backed platforms get all three from the scheduler itself, which is why this gap never closes no matter how many templates ship.
Declarative machine provisioning. All three tools start from servers you already have: install an agent or run a script on a VPS, then register it with the panel. There is no MachineDeployment-style object that declares "this fleet has six workers" and reconciles reality toward it — no auditable, git-storable machine lifecycle, no automatic replacement of a failed node, no declarative scale-up. Coolify Cloud's per-server pricing ($5 base for two servers, +$3 per extra) makes the economics of this visible: every server is a pet you adopted individually, and the bill itemizes exactly how many pets you have. A Cluster API fleet inverts the model — machines are cattle declared in manifests, and adding capacity is a pull request, not a provisioning ticket.
An API surface agents can actually drive. This is the gap the 2026 roundups mention least and working platform teams feel first. Coolify has the strongest story — an OpenAPI-described API, a Terraform provider managing dozens of resource types, and a community MCP server exposing 60+ tools for deploy-and-diagnose workflows. Dokploy has its own API plus community agent skills. CapRover has a CLI and an API, but agent-ecosystem energy around it is visibly lower.
Still, in all three cases the machine-readable surface is a layer on top of a dashboard-first design — not the platform's native contract the way a Render-compatible API is for an API-first PaaS. As deploy-from-chat workflows move from demo to daily driver, "can an agent provision, deploy, roll back, and read fleet health without screen-scraping a UI" stops being a nice-to-have. It is the one comparison axis where none of the three leaders was designed API-first.
Which One to Pick, and When You Have Outgrown All Three
The honest verdict is situational, and the scoreboard above makes it easy to state:
- Solo developer, side projects, one VPS: pick whichever UI you prefer — Coolify for the biggest community and template ecosystem, Dokploy for the cleaner Compose-native feel. Both get you from zero to HTTPS in minutes, and at this scale their ceilings do not matter.
- Small team, client work, a handful of servers: Coolify's stable v4 plus its managed cloud panel (someone else runs the brain while your servers stay the muscle) is the lowest-ops path; Dokploy's pre-1.0 status deserves a harder look before you bet client uptime on it.
- "Proven and boring" with minimal churn: CapRover. The dated UI is real, Compose support is weaker, but nine years of production surface and a 2026 security hotfix mean it changes under you the least.
And the signal you have outgrown all three at once: the week you need a second machine to be a scheduling decision instead of a provisioning project — when workloads must survive a node loss rather than be restarted after one, when tenants must not share a Docker daemon's blast radius, or when an agent needs a Render-compatible API rather than a dashboard with an API beside it. That is the exact seam where template counts stop helping and fleet lifecycle starts: declarative machines, a real scheduler, storage and identity that follow workloads across nodes.
Winning on DX templates was always a different bet than winning on fleet lifecycle. The star counts told you who won the first bet. The second bet is still open.
Bex.co is the open-source, AI-native Render alternative — push a git repo, get a running HTTPS service on machines you own, with a Render-compatible API your agents can drive from day one instead of a dashboard they have to work around. Star the repo on GitHub or deploy your first app today.
