필사 모드: DevOps PaaS Platforms 2026 — Vercel / Fly.io / Render / Railway / Coolify / Dokploy / Kamal Deep Dive

Prologue — Heroku left, and 14 PaaS platforms grew in its place

When Heroku unveiled "git push to deploy" in 2007, it looked like magic. One line — `git push heroku main` — sent your Rails app to the internet, and that became the definition of PaaS.

Then in November 2022, Salesforce-owned Heroku killed the free tier. That single event shaped the 2026 PaaS landscape. The developers who had built side projects on Heroku scattered, and 14 PaaS platforms grew to fill the space.

This is the picture as of May 2026:

- **Vercel** became the home of Next.js, and Fluid Compute filled the gaps of serverless.

- **Fly.io** runs containers close to bare metal across multiple regions.

- **Render** is the most direct Heroku replacement.

- **Railway** is beloved for usability and raised Series B again in 2024.

- **Cloudflare** completed the full-stack edge with Pages, Workers, and R2.

- **Netlify** evolved from JAMstack pioneer to composable full-stack platform.

- **Coolify** became the de facto standard for open-source self-hosted PaaS, with seed funding in 2024.

- **Dokploy** challenges Coolify.

- **Kamal 2** (37signals) revived the old virtue of deploying to your own servers.

- **Heroku** survived even past the rumored 2025 IBM acquisition.

- **Northflank** wraps Kubernetes power in PaaS UX.

- **Sevalla** is the new PaaS from Kinsta.

- **Caprover / Easypanel** keep Docker-first self-hosting alive.

This piece compares all 14 in one sitting and answers — who should pick what.

1. The 2026 PaaS map — managed / open-source self-hosted / roll-your-own on a VPS

In one line, a PaaS is "you git push your code and it runs on the internet." But in 2026, that single line splits into three lanes.

1.1 Three axes

**Axis 1 — Managed cloud PaaS.** Vercel, Fly, Render, Railway, Cloudflare, Netlify, Heroku, Northflank, Sevalla. Enter a card and you stop worrying. Infrastructure becomes invisible. In return, monthly usage shows up on the invoice.

**Axis 2 — Open-source self-hosted PaaS.** Coolify, Dokploy, Caprover, Easypanel. Install on a single VPS (Hetzner, DigitalOcean, OVH, Vultr) and you get a Vercel-like experience on top of it. Infrastructure cost is just the VPS bill. Backups, monitoring, security patches are your job.

**Axis 3 — Roll your own on a VPS / lightweight container orchestrator.** Kamal 2. Not a PaaS but "a clean way to deploy containers to servers you chose." Built by 37signals under DHH's hand, and the default deploy tool for Rails 7.2.

1.2 Cost vs control matrix

CONTROL ↑

|

Kamal 2 ────────|──────── Coolify / Dokploy

(raw VPS ops) | (OSS self-hosted)

|

────────────────+──────────────── COST →

|

Heroku / Render | Vercel / Cloudflare

(classic PaaS) | (edge / serverless)

|

CONTROL ↓

Horizontal is monthly cost, vertical is infrastructure control. **Upper left** is cheap but you do everything. **Lower right** is one card and everything runs.

After the 2022 Heroku free-tier shutdown, the center of gravity for side projects **shifted to the upper left.** A USD-5 Hetzner VPS with Coolify installed can run five projects at once. That is the new default in 2026.

1.3 Git-driven deploys — the standard everyone shares

There is one thing every platform on all three axes shares. **Git push is deploy.**

- Push to `main` and production updates.

- Open a PR and a preview URL appears.

- Close the PR and the preview disappears.

Vercel standardized this flow in 2017, and by 2026 it reached even self-hosted PaaS defaults. Coolify, Dokploy, and Caprover all support GitHub webhooks and have preview-environment features.

2. Vercel — home of Next.js, inventor of Fluid Compute

In 2026 Vercel does two things at once. It is the home of Next.js, and with Fluid Compute it redefined the serverless paradigm.

2.1 Fluid Compute — the model that ended cold starts

Announced in August 2024, **Fluid Compute** can be summarized as "one instance handles many concurrent requests."

The classic AWS Lambda / Vercel Functions model:

- 1 request → 1 instance.

- 100 concurrent requests → 100 instances.

- Cold starts possible on every instance.

The Fluid Compute model:

- 1 instance handles N concurrent requests.

- While waiting on an AI response, yield CPU to other requests.

- When a cold start happens, its cost amortizes across N requests.

It especially shines for **AI/LLM backends**. If you call ChatGPT/Anthropic APIs and spend 90% of your time waiting on the network, other requests can use that idle time. Vercel claims about 1/3 the cost for the same traffic, and benchmarks match.

2.2 Vercel's strengths

- **1:1 Next.js integration.** App Router, React Server Components, Server Actions, Streaming, and PPR (Partial Prerendering) ship first and work best.

- **Preview Deployments.** Every PR gets a unique URL. Clients leaving feedback on PR comments has become standard.

- **Edge Network.** 100+ POPs. Static assets serve from a nearby location.

- **Vercel KV / Postgres / Blob.** A managed data tier built on top of Upstash, Neon, R2.

- **AI SDK.** The most polished TypeScript library for LLMs.

2.3 Vercel's weaknesses

- **Price.** Costs scale fast with traffic. Especially when you have heavy image optimization or many Edge Middleware requests.

- **Vendor lock-in.** Next.js itself is OSS, but ISR, Image Optimization, and Edge Config are hard to reproduce outside Vercel.

- **Bad fit for long-running backends.** Even with Fluid Compute, single-request time limits exist (60s–300s on Pro). WebSocket and gRPC streaming servers do not fit.

2.4 Who should pick Vercel

- Next.js SaaS, marketing sites, dashboards.

- Next.js apps with LLM backends like AI chatbots and search.

- Agencies that collaborate with clients per-PR.

Avoid if: you run 24/7 daemons, serve 5 GB ML models, or have a strict cost ceiling on a side project.

3. Fly.io — multi-region + bare-metal friendly

Fly.io's one-line summary is "your Docker image runs in 30 cities at once." What differs from Vercel and Render is **Fly Machines** — Firecracker microVM-based containers running close to physical servers.

3.1 Multi-region model

When you deploy an app on Fly, you list regions in `fly.toml`.

app = "my-app"

primary_region = "nrt"

[build]

image = "ghcr.io/me/app:latest"

[[regions]]

code = "nrt"

[[regions]]

code = "iad"

[[regions]]

code = "fra"

This single setting starts instances in Tokyo, Virginia, and Frankfurt at once. Traffic routes to the nearest region. If you need Postgres, `fly pg create` spins up multi-region Postgres with WAL replication.

3.2 Fly's strengths

- **Near bare-metal performance.** Firecracker microVMs add only one thin layer over containers.

- **30+ regions.** Smaller than AWS/GCP, but dominant in the PaaS category.

- **GPU support (added 2024).** A100, L40S, H100 by the hour.

- **WireGuard-based private network.** Instances talk directly over IPv6.

- **Reasonable pricing.** Small instances start at USD 2–5/mo.

3.3 Fly's weaknesses

- **More ops burden than other PaaS.** Fly Postgres is not a managed DB but "Stolon Postgres that Fly manages for you." User reports show you still need to monitor backups yourself.

- **CLI-first over the dashboard.** `flyctl` is central, the dashboard is secondary.

- **Billing history.** A 2023 billing policy change caused issues for some users. It took time to rebuild trust.

3.4 Who should pick Fly

- SaaS with a global user base.

- Workloads where latency is critical — game backends, realtime communication servers.

- Small AI services that need GPUs (Cog, BentoML containers).

- Teams that already have a Docker image and want infra control.

4. Render — the right answer for Heroku replacement

In 2026 Render is the **"most Heroku-like new PaaS."** Founded 2019, Series B 50M USD in 2022, Series C in 2025 with serious enterprise moves.

4.1 Render's model

- **Web Service.** A container that takes HTTP traffic.

- **Background Worker.** Queue workers like Sidekiq, Celery, BullMQ.

- **Cron Job.** Scheduled tasks.

- **Static Site.** SPA and static site hosting.

- **Private Service.** Internal services with no external exposure.

- **PostgreSQL / Redis / Key Value.** Managed data tier.

A single `render.yaml` (Render Blueprint) declares the whole infrastructure as code. It looks like an evolution of Heroku's `app.json`.

services:

- type: web

name: api

runtime: docker

dockerfilePath: ./Dockerfile

envVars:

- key: DATABASE_URL

fromDatabase:

name: api-db

property: connectionString

databases:

- name: api-db

databaseName: api

plan: standard

4.2 Render's strengths

- **PaaS-feel.** As smooth as Vercel, more modern than Heroku.

- **Official managed Postgres / Redis.** Backups, HA, encryption by default.

- **Preview Environment.** Full-stack PR preview including the database.

- **Reasonable pricing.** Web Service Starter from USD 7/mo, free for static sites.

4.3 Render's weaknesses

- **Few regions.** About 5 (Oregon, Ohio, Frankfurt, Singapore, etc.) — fewer than Vercel or Fly.

- **Cold starts.** Free and Starter Web Services sleep when idle.

- **Basic logs and metrics.** Datadog or Grafana integration is on you.

4.4 Who should pick Render

- Traditional monolithic backends — Rails, Django, Express.

- Teams migrating off Heroku.

- Small-to-mid SaaS that needs both dashboard and automation.

5. Railway — usability champion, Series B re-raised in 2024

Railway is loved for usability alone. They raised a Series B of 20M USD in May 2024 — proof that "PaaS is not dead."

5.1 Railway's model

- Everything is a "service." Postgres, Redis, MongoDB, RabbitMQ — all services.

- Drag-and-drop services on a GUI to connect them.

- Environment variables reference other services (pull `POSTGRES_URL` directly from the Postgres service).

- **Nixpacks** (Railway's Buildpacks successor) builds most languages without a Dockerfile.

5.2 Railway's strengths

- **Smoothest setup.** Connect GitHub → pick repo → done.

- **Full stack on one screen.** Backend, DB, workers, crons in one board.

- **Rich official templates.** Plausible, n8n, Outline, Ghost — one-click deploy.

- **Usage-based pricing.** Don't pay for idle.

5.3 Railway's weaknesses

- **Few regions.** US-West, US-East, EU, Asia. Fewer than Vercel and Fly.

- **Per-service caps.** vCPU and memory ceilings are lower than managed clouds.

- **Billing is usage-based — hard to predict.** Traffic spikes jump the bill.

5.4 Who should pick Railway

- Side projects, MVPs, hackathons.

- Solo developers who want to spin up a full stack fast.

- Discord bots, Telegram bots, automation hosting.

6. Cloudflare Pages + Workers + R2 — full-stack edge complete

Cloudflare entered the PaaS market last and with the most different model. The 2026 Cloudflare stack deserves the name **full-stack edge.**

6.1 Cloudflare stack composition

- **Pages** — static and SPA hosting + Functions (serverless).

- **Workers** — edge V8 isolates. Running on 300+ POPs worldwide.

- **R2** — S3-compatible object storage. **Egress is free** — the killer feature.

- **D1** — SQLite at the edge, with read replicas globally.

- **KV / Durable Objects / Queues / Vectorize** — a full set of edge data primitives.

6.2 Cloudflare's strengths

- **Zero egress cost.** R2 broke S3's pricing model.

- **Global edge.** 300+ POPs. Close to anyone.

- **Almost no cold starts.** V8 isolates spin up in milliseconds.

- **Generous free tier.** Workers 100k requests/day free, Pages unlimited static.

6.3 Cloudflare's weaknesses

- **Partial Node.js compatibility.** Modules like `node:fs` and `node:net` are restricted (the `nodejs_compat` flag covers part of it).

- **Single-request CPU time limit.** Free 50ms, Paid 30s. Heavy synchronous work is a bad fit.

- **D1 doesn't yet support global write distribution.** Primary region + global read replicas model.

- **Vendor lock-in.** Durable Objects, KV, Queues are not reproducible outside Cloudflare.

6.4 Who should pick Cloudflare

- Sites where static assets dominate traffic — blogs, docs, landing pages.

- Edge-friendly apps built with Astro, Remix, Hono, or Next.js (via adapter).

- Image and video hosting services (R2 egress is free).

- Workloads where global users and minimal cold starts matter.

7. Netlify — JAMstack pioneer, transformed into composable full-stack

Netlify is the company that coined JAMstack in 2014. The 2026 Netlify adds full-stack capabilities to that identity.

7.1 Netlify today

- **Netlify Edge Functions.** Deno-based. An edge runtime similar to Cloudflare Workers.

- **Functions.** Serverless on top of AWS Lambda.

- **Forms.** Form submissions plus spam filtering built in.

- **Identity / Authentication.** Built on GoTrue.

- **Composable Web.** A platform concept that ties Headless CMS, Commerce, Auth, and Search together.

7.2 Netlify's strengths

- **Standard for static sites.** Jekyll, Hugo, Astro, Gatsby, and Next.js Export all work smoothly.

- **Build Plugins.** Extend the build pipeline as npm packages.

- **Clear focus.** Marketing sites, blogs, docs, and e-commerce frontends.

7.3 Netlify's weaknesses

- **Slower Next.js new-feature support than Vercel.** PPR and Streaming RSC lag 1–2 quarters.

- **Bad for long-running backends.** Function timeouts are short.

- **Bills rise fast.** Build time grows the invoice faster than traffic.

7.4 Who should pick Netlify

- Marketing, blog, and documentation sites.

- Teams using meta-frameworks other than Next.js — Astro, Eleventy, Hugo.

- Non-technical operators who need built-in forms and identity.

8. Coolify (open source) — self-hosted Vercel

Coolify became the de facto standard for open-source self-hosted PaaS in 2026. A 2024 seed round kicked off full-time development, and GitHub stars passed 40k.

8.1 Coolify's model

- Install on a VPS with a one-line script.

- Connect GitHub or GitLab via OAuth.

- Pick a repo and it auto-detects Nixpacks, Dockerfile, or Compose.

- Open a PR and get an auto-generated preview URL.

- Spin up Postgres, MySQL, Redis, MongoDB via GUI click.

- Traefik handles automatic SSL via Let's Encrypt and reverse proxy.

8.2 Coolify's strengths

- **Fully self-hosted.** Install on a Hetzner CX22 (EUR 4/mo) and run five projects comfortably.

- **Smooth UX.** They're proud of copying Vercel.

- **Official templates.** Plausible, Umami, Outline, Ghost, n8n — one-click deploy.

- **Data sovereignty.** Your DB lives on your VPS. Helpful for GDPR and corporate policies.

8.3 Coolify's weaknesses

- **You own operations.** Disk-full, OOM, SSL expiry are yours to watch.

- **HA is hard.** One node dies, all services die. Multi-node is in beta.

- **Backups are separate.** You set up a cron to `pg_dumpall` to S3 yourself.

8.4 Who should pick Coolify

- Developers running 5–20 side projects on one VPS.

- Teams blocked from managed cloud by GDPR, HIPAA, or local policy.

- Small startups saving USD 100–200/mo.

9. Dokploy (open source) — Coolify alternative

Dokploy appeared in 2024 as an alternative to Coolify. As of May 2026, GitHub stars passed 20k, and it attracts "people who thought Coolify felt heavy."

9.1 Dokploy vs Coolify

| Item | Coolify | Dokploy |

|------|---------|---------|

| First appeared | 2020 | 2024 |

| GitHub stars (May 2026) | 40k+ | 20k+ |

| Core language | PHP (Laravel) | TypeScript (Next.js) |

| Database | Postgres | Postgres |

| Reverse proxy | Traefik | Traefik |

| Multi-server | Beta | Stable |

| Pricing | Free / Cloud | Free / Cloud |

Dokploy's differentiator is that multi-server support shipped as a stable feature. Coolify still has multi-node in beta.

9.2 Who should pick Dokploy

- Teams comfortable with TypeScript (if you'd consider contributing).

- Teams that want to spread workloads across multiple servers.

- People who tried Coolify and want something lighter.

10. Kamal 2 (37signals) — deploy to your own servers

Kamal is not a PaaS. It is a **container deploy tool.** 37signals (Basecamp, Hey) built it as part of leaving AWS, and Kamal 2 shipped in 2024.

10.1 Kamal's philosophy

- Install containers on servers you bought — Hetzner, OVH, your own datacenter.

- No PaaS abstraction. You know the server, domain, and SSL.

- `kamal deploy` for zero-downtime deploys and rollbacks.

- Uses Kamal-proxy (a small in-house proxy) instead of Traefik.

- The default deploy tool for Rails 7.2.

10.2 Example `deploy.yml`

service: hey

image: 37signals/hey

servers:

web:

- 192.168.0.1

- 192.168.0.2

job:

hosts:

- 192.168.0.3

cmd: bin/jobs

registry:

server: ghcr.io

username: my-org

password:

- KAMAL_REGISTRY_PASSWORD

env:

secret:

- RAILS_MASTER_KEY

- DATABASE_URL

This single file enables `kamal setup` → `kamal deploy`.

10.3 Kamal's strengths

- **Cheapest ops cost.** At similar traffic, roughly 1/10 of AWS and 1/5 of Heroku.

- **Zero vendor lock-in.** Any cloud where SSH works is fine.

- **DHH philosophy proven.** Basecamp and Hey actually run on this.

- **2025 ONCE / Campfire** — self-hosted SaaS running on Kamal became a validated model.

10.4 Kamal's weaknesses

- **No PaaS UX.** No preview environments, no auto DB provisioning.

- **You actually operate the server.** Security patches, firewall, monitoring.

- **You design HA yourself.** No auto-failover.

10.5 Who should pick Kamal

- Teams running monolithic Rails, Django, Laravel.

- Teams already spending USD 1000+/mo on AWS or Heroku.

- Enterprises where infrastructure control is a policy requirement.

11. Northflank / Sevalla / Caprover / Easypanel — other PaaS

11.1 Northflank — peak of Kubernetes PaaS

Northflank is the best at "wrapping Kubernetes in a PaaS UX." In 2026 it is one of the managed PaaS that takes the heaviest workloads.

- Deploys on top of multiple clouds (AWS, GCP, Azure).

- Stable GPU workload support.

- Microservice topology graph.

- BYOC (Bring Your Own Cloud) into your own cloud account.

Target: teams with 30+ containers, complex microservices, and enterprise compliance needs.

11.2 Sevalla — Kinsta's new PaaS

Sevalla is the new PaaS Kinsta — famous for managed WordPress hosting — launched in 2024. Built on Google Cloud, aimed at "the gaps in Vercel and Render."

- 30+ GCP regions globally.

- Static sites, apps, DBs, object storage in one platform.

- 100 free static sites, reasonable app pricing.

Target: agencies running PHP and Node broadly — WordPress, Laravel, and friends.

11.3 Caprover — the oldest self-hosted PaaS

Caprover appeared in 2017. Docker Swarm-based. Before Coolify, it was the standard for self-hosted PaaS. Still 13k+ GitHub stars and stable.

- Docker Swarm.

- 200+ one-click apps.

- Caprover-CLI for CI/CD integration.

Target: long-time users and teams that deliberately use Docker Swarm.

11.4 Easypanel — lightweight self-hosting

Easypanel appeared in 2022. Docker Compose-based. Lighter than Coolify with a clean UI.

- Docker Compose.

- 100+ templates.

- Free / Pro (USD 19/mo).

Target: solo developers running 3–5 apps cleanly on a single VPS.

12. Heroku — a giant past its era

Heroku is still alive in 2026. It lost users after the 2022 free-tier shutdown but runs on the enterprise Salesforce customer base. In 2025 reports surfaced that IBM was considering acquiring Heroku, but as of May 2026 nothing is confirmed.

12.1 Heroku today

- **Mini Dyno** at USD 5/mo replaces the free tier.

- **Heroku Postgres / Redis** still the standard managed data tier.

- **Add-on marketplace** is alive (Logentries, Papertrail, NewRelic).

- **Fir Generation Platform**, GA'd in 2024 — a container-based next-gen runtime.

12.2 What Heroku still offers

- Stability. It has run for 17 years.

- If you already have a Heroku app, maintenance beats migration cost.

- Useful for enterprises with Salesforce integration.

12.3 Starting fresh?

Putting a new side project on Heroku is nostalgia. Render, Railway, and Coolify do the same thing cheaper and better. But if work already uses Heroku, the previous decision is not necessarily wrong.

13. Korea & Japan — Kakao i Cloud, Naver NCP, Sakura, KAGOYA, Conoha

Global PaaS isn't the whole story. Korea and Japan have their own PaaS and VPS ecosystems.

13.1 Korea — Kakao i Cloud / Naver NCP

- **Kakao i Cloud.** Kakao's cloud. Launched 2022. Strong on Korean data sovereignty. PaaS abstraction is closer to AWS.

- **Naver Cloud Platform (NCP).** Naver's cloud. Strong in Korean government, public sector, and finance. Container and Kubernetes services are stable.

- **NHN Cloud.** Built on operations know-how from Hangame and PAYCO. Friendly to Korean enterprises.

If a Korean startup uses managed PaaS, Vercel, Render, or Fly come first when the user base is global. Naver NCP and Kakao i Cloud are more efficient when users are inside Korea.

13.2 Japan — Sakura, KAGOYA, Conoha, Heroku Japan

- **Sakura Cloud.** Japan's own cloud. Datacenters inside Japan, friendly to government and finance.

- **KAGOYA CLOUD.** Strong in the VPS market. Low-cost VPS from JPY 590/mo.

- **ConoHa.** Part of GMO. Famous for WordPress and Minecraft hosting.

- **Heroku Japan.** Under Salesforce Japan. Japanese support plus Tokyo region.

The 2026 standard for Japanese developer side projects has become "KAGOYA or Conoha VPS + Coolify or Dokploy."

14. Who should pick what — side project / startup / global / cost-sensitive

14.1 Side projects

- **Start at USD 0.** Cloudflare Pages + Workers + R2. The most generous free tier.

- **Under USD 5.** Hetzner CX22 + Coolify. Five to ten projects at once.

- **Only Next.js.** Vercel Hobby. Free with traffic caps.

- **Discord bots / automation.** Railway, usage-based.

14.2 Early-stage startup (seed)

- **Fast market entry.** Vercel + Supabase or Neon + Resend. Forget infrastructure.

- **Monolithic Rails / Django.** Render. Heroku-like reliability.

- **Global users.** Fly.io + Postgres. Multi-region from day one.

- **Cost-sensitive.** Hetzner + Coolify + Cloudflare CDN. Survive to Series A on USD 20/mo.

14.3 Series A–B startup

- **Pick one of Vercel / Fly / Render + a separate managed DB** (Neon, Supabase, PlanetScale).

- **Split observability out.** Datadog, Grafana Cloud, Axiom, Better Stack.

- **CDN/WAF goes to Cloudflare.** The two-layer model is now standard.

14.4 Global users / edge

- **Static-heavy.** Cloudflare Pages + R2.

- **Dynamic API.** Fly.io multi-region or Vercel Edge Functions.

- **DB.** Neon (serverless Postgres) / Turso (edge SQLite) / PlanetScale.

14.5 Cost-sensitive (solo / bootstrap)

- **Hetzner CX22 (EUR 4/mo) + Coolify** is effectively the answer.

- **Free Cloudflare CDN / WAF.**

- **Cloudflare R2** for object storage (egress free).

- **Backups** via GitHub Actions cron — `pg_dumpall` → R2.

This combo runs 5–20 projects in under USD 10/mo. The 2026 bootstrapper standard.

14.6 Korea / Japan local users

- Korea: Naver NCP + an own managed DB.

- Japan: Sakura Cloud + Sakura Object Storage.

- Side projects: KAGOYA / Conoha VPS + Coolify.

15. Decision tree — pick in 30 seconds

Side project? USD 0–5/mo?

├── Only Next.js → Vercel Hobby

├── Full stack + DB → Railway (usage-based)

├── 5+ projects → Hetzner + Coolify

└── Static + edge → Cloudflare Pages

Startup MVP? USD 20–100/mo?

├── Next.js → Vercel Pro + Neon

├── Rails/Django → Render

├── Multi-region → Fly.io

└── Cost-first → Coolify on Hetzner + Cloudflare

Growth stage? USD 500+/mo?

├── Keep going, start splitting

├── DB to managed external (Neon / Supabase / PlanetScale)

├── Observability to Datadog / Grafana / Axiom

└── CDN/WAF to Cloudflare

Enterprise?

├── BYOC → Northflank

├── Own datacenter → Kamal 2 + own K8s

└── Compliance → AWS/Azure/GCP direct

References

- Vercel — [Fluid Compute announcement (Aug 2024)](https://vercel.com/blog/introducing-fluid-compute)

- Vercel — [Next.js 15 App Router](https://nextjs.org/docs/app)

- Fly.io — [Fly Machines](https://fly.io/docs/machines/)

- Fly.io — [GPU on Fly](https://fly.io/docs/gpus/)

- Render — [Render Blueprints (render.yaml)](https://render.com/docs/blueprint-spec)

- Render — [Preview Environments](https://render.com/docs/preview-environments)

- Railway — [Series B 2024 (TechCrunch)](https://techcrunch.com/2024/05/01/railway-raises-20-million-paas/)

- Railway — [Nixpacks](https://nixpacks.com/)

- Cloudflare — [Pages](https://pages.cloudflare.com/)

- Cloudflare — [Workers Runtime APIs](https://developers.cloudflare.com/workers/runtime-apis/)

- Cloudflare — [R2 (egress-free storage)](https://developers.cloudflare.com/r2/)

- Cloudflare — [D1 SQLite](https://developers.cloudflare.com/d1/)

- Netlify — [Edge Functions](https://docs.netlify.com/edge-functions/overview/)

- Netlify — [Composable Web](https://www.netlify.com/composable-web/)

- Coolify — [Official site](https://coolify.io/)

- Coolify — [GitHub repo](https://github.com/coollabsio/coolify)

- Dokploy — [Official site](https://dokploy.com/)

- Dokploy — [GitHub repo](https://github.com/dokploy/dokploy)

- Kamal — [Official site](https://kamal-deploy.org/)

- Kamal — [Kamal 2 announcement (37signals 2024)](https://world.hey.com/dhh/kamal-2-1ad0d4ef)

- 37signals — [Leaving the cloud (2023)](https://world.hey.com/dhh/why-we-re-leaving-the-cloud-654b47e0)

- Heroku — [Fir Generation Platform](https://blog.heroku.com/fir-generation-platform-now-available)

- Heroku — [Pricing (post-free 2022)](https://www.heroku.com/pricing)

- Northflank — [Official site](https://northflank.com/)

- Sevalla — [Official site (Kinsta)](https://sevalla.com/)

- Caprover — [Official site](https://caprover.com/)

- Easypanel — [Official site](https://easypanel.io/)

- Naver Cloud Platform — [Official site](https://www.ncloud.com/)

- Kakao i Cloud — [Official site](https://www.kakaocloud.com/)

- Sakura Internet — [Official site](https://www.sakura.ad.jp/)

- KAGOYA Cloud — [Official site](https://www.kagoya.jp/cloud/)

- ConoHa — [Official site](https://www.conoha.jp/)

- Hetzner — [Cloud pricing](https://www.hetzner.com/cloud/)