Hosting a small Node web app from Git: Render, Railway, Sevalla and Fly.io
Render runs a small Node app and its database for $13 a month at fixed instance prices, with database point-in-time recovery. Its lead is narrow; Railway suits teams that want usage billing and can accept its May 2026 outage record.
Render
Fixed $7 instance and $6 Postgres, database point-in-time recovery and rollbacks to a retained build, with no workspace fee on Hobby.
Our lead is narrow; Hobby is described for personal projects, and request logs or more than one seat need Pro at $25 a month.
Why Render is our PickTeam behind Render? Get your award badge
No Rising Star selected
Zeabur and Leapcell were publicly available before the two-year cutoff. Sevalla continues Kinsta hosting, while the container and function alternatives come from established platform owners.
Some launch directories could not be read, so this does not establish that no suitable newcomer exists.
Who is this article for? Developers putting one small server-rendered Node web app and its Postgres database into production from Git, on a platform that runs the servers for them.
About 23 min read
What Preferred Pick, Rising Star and Ready to Rock mean
The quick list
Choose Render for one person running an always-on Node web service and the smallest paid Postgres database. Compare the monthly price, then how each platform recovers your code and data.
Normal monthly USD prices for one always-on service with about 512 MB of memory and the smallest paid managed Postgres, pricing checked October 30, 2026. Our arithmetic, not quotes. Railway bills by usage above its plan minimum. Fly.io pricing checked October 30: about $42.69/month includes 1 GB used database storage and 1 GB North America/Europe outbound traffic. Sevalla is about $17.10/month with 100 build minutes and 1 GB external traffic. Prices for several other suitable platforms could not be confirmed. Excludes tax, bandwidth and build minutes beyond allowances, extra storage and add-ons.
The comparison at a glance
Compare the monthly price, plan boundaries, spending controls and what each platform restores after a bad deploy or a lost disk.
Render is our recommendation for one person running a small production app. Its lead is narrow; Sevalla offers a managed Git workflow and Fly.io offers more deployment control. All tools remain selectable.
Hides matching answers for the selected tools. Unknowns and unverified claims stay visible.
| Feature | Render | Railway | Sevalla | Fly.io |
|---|---|---|---|---|
| Best fit | Preferred Pick: one person running a small production Node app | Several people need access; usage billing with a hard spending limit | Managed Git deployment with app and database pricing | Managed database recovery and container deployment control |
| One always-on app plus smallest paid Postgres | $13/month on Hobby ($7 512 MB instance + $6 Postgres); $38 on Pro | Usage-metered: Pro $20 minimum includes $20 of usage; Hobby $5 minimum for side projects | About $17.10/month: $10 S1 + $5 Database 1 + $2 builds + $0.10 egress | About $42.69/month: $4.39 Machine + $38 managed Postgres + $0.28 storage + $0.02 egress |
| Workspace or account plan | Hobby $0: one seat, “personal projects and prototypes”; Pro $25 flat, unlimited seats | Hobby: single developer, side projects; Pro: unlimited seats, production teams | Kinsta-managed app/database resources; no seat charge in this example | Pay-as-you-go infrastructure; optional paid support excluded |
| Spending control | Limit covers build minutes only; instance prices fixed | Email alerts; hard limit makes workloads unreachable | Builds $0.02/minute; external traffic $0.10/GB; spending cap not established | Overages billed without a ceiling; no billing alerts yet |
| Managed Postgres recovery | Point-in-time recovery: 3 days on Hobby, 7 on Pro | Point-in-time recovery, about 4 weeks after enabling; Pro per support answer | Daily backups advertised; entry database restore window not established | Managed backups and high availability; point-in-time restore documented August 2026 |
| Persistent application files | Paid disk $0.25/GB; daily snapshots kept 7+ days; one instance only | Volume mounted at start; no replicas; brief downtime on redeploy | Optional persistent storage starts at $3/month for 10 GB; excluded here | Optional volumes priced separately; no app volume in this example |
| Rollback scope | Retained build and settings; 5 builds on Hobby, 15 on Pro; disk unchanged | Previous image and custom variables; 72 h image retention on Hobby, 120 h on Pro | Earlier image and build-time variables; current runtime settings retained; database unchanged | Deployment controls documented; app rollback does not replace database restore |
| 2026 reliability record | Frequent build and regional incidents; one critical 2-hour Oregon service disruption (July 24) | Own report: about 8 hours on May 19–20, all workloads unreachable at peak | Comparable 2026 incident sample not established | Comparable 2026 incident sample not established |
Managed Git-to-server hosting for one small Node web app. Operational sources checked October 19, 2026; Render, Railway and Heroku prices and Sevalla/Fly.io sources checked October 30, 2026. Items marked as read September 7 were not reread. Next review December 18, 2026, or earlier after a material change. Example totals assume one always-on service with about 512 MB of memory, the smallest paid managed Postgres and one person with access, at normal monthly USD rates before tax; they exclude bandwidth and build minutes beyond allowances, extra storage and add-ons. Prices and required configuration details for several other suitable platforms could not be confirmed. Render remains our recommendation, with a narrow lead. This is public-source research: documentation does not establish deployment success, uptime or security in practice. See the evidence section for user reports and incident records.
All features and sources 15 criteria
Open the groups that matter to you. They compare the same selected tools. “Planned” means future work, not an available feature. Sources include the published conditions.
Price, plan and operation
Show 8 criteria
| Feature | Render | Railway | Sevalla | Fly.io |
|---|---|---|---|---|
| Best fit | Preferred Pick: one person running a small production Node app | Several people need access; usage billing with a hard spending limit | Managed Git deployment with app and database pricing | Managed database recovery and container deployment control |
| One always-on app plus smallest paid Postgres | $13/month on Hobby ($7 512 MB instance + $6 Postgres); $38 on Pro | Usage-metered: Pro $20 minimum includes $20 of usage; Hobby $5 minimum for side projects | About $17.10/month: $10 S1 + $5 Database 1 + $2 builds + $0.10 egress | About $42.69/month: $4.39 Machine + $38 managed Postgres + $0.28 storage + $0.02 egress |
| Workspace or account plan | Hobby $0: one seat, “personal projects and prototypes”; Pro $25 flat, unlimited seats | Hobby: single developer, side projects; Pro: unlimited seats, production teams | Kinsta-managed app/database resources; no seat charge in this example | Pay-as-you-go infrastructure; optional paid support excluded |
| Git build and runtime | Linked branch; native Node or Docker | GitHub or Dockerfile; Express guide | Git repository builds and automatic deployments | Git-connected deployment; container configuration in fly.toml |
| Idle behavior | Paid instances stay on; Free sleeps after 15 minutes | Optional sleep after 5–10 minutes without outbound traffic | Persistent runtime; optional hibernation excluded | Keep one Machine running with autostop disabled |
| Spending control | Limit covers build minutes only; instance prices fixed | Email alerts; hard limit makes workloads unreachable | Builds $0.02/minute; external traffic $0.10/GB; spending cap not established | Overages billed without a ceiling; no billing alerts yet |
| Logs | 7 days on Hobby, 14 on Pro; HTTP request logs Pro+ | 7 days on Hobby, 30 on Pro; 500 lines/second cap | Deployment build/runtime logs; comparable retention allowance not established | Metrics and support available; comparable log-retention allowance not established |
| Custom domains | 2 on Hobby, 15 on Pro, then $0.25/month each | Custom domains; automatic certificates | S1 supports custom domains; cheaper H1 does not | Shared IPv4 and automatic TLS; dedicated IPv4 optional |
These selected tools have matching answers in this group. Turn off “Show differences only” to see them.
Sources for price, plan and operation
Sevalla app pricing, database pricing, deployments, terms and privacy, checked October 30, 2026. Example includes 100 build minutes and 1 GB external traffic. Fly.io pricing, Managed Postgres, cost management, changelog and terms, checked October 30, 2026. Example assumes one always-on Machine, 1 GB used database storage and 1 GB North America/Europe outbound traffic, before tax. Render, Railway and Heroku pricing rechecked October 30, 2026; other operational documentation checked October 19, 2026: Render pricing, plan change, web services, free tier, billing FAQ, logging; Railway plans, pricing page, Express, sleep, cost control, logs, staff answer on commercial use; Heroku pricing, Node, Eco, billing, logs; Koyeb pricing, build and sleep, read September 7. Totals are our arithmetic, not quotes; taxes, bandwidth and extra resources can add charges.
Recovery, data and continuity
Show 7 criteria
| Feature | Render | Railway | Sevalla | Fly.io |
|---|---|---|---|---|
| Managed Postgres recovery | Point-in-time recovery: 3 days on Hobby, 7 on Pro | Point-in-time recovery, about 4 weeks after enabling; Pro per support answer | Daily backups advertised; entry database restore window not established | Managed backups and high availability; point-in-time restore documented August 2026 |
| Persistent application files | Paid disk $0.25/GB; daily snapshots kept 7+ days; one instance only | Volume mounted at start; no replicas; brief downtime on redeploy | Optional persistent storage starts at $3/month for 10 GB; excluded here | Optional volumes priced separately; no app volume in this example |
| Rollback scope | Retained build and settings; 5 builds on Hobby, 15 on Pro; disk unchanged | Previous image and custom variables; 72 h image retention on Hobby, 120 h on Pro | Earlier image and build-time variables; current runtime settings retained; database unchanged | Deployment controls documented; app rollback does not replace database restore |
| Runtime configuration | Environment variables and secret files | Variables; sealed values cannot be read back or copied | Current runtime environment retained on code rollback | Keep app configuration and credentials separately for migration |
| Exit preparation | Copy disk and database; keep source and configuration | Download volume files by CLI; keep sealed values separately | Keep source, configuration and independent Postgres exports; full export procedure unestablished | Keep source and fly.toml; database migration via pg_dump and pg_restore |
| 2026 reliability record | Frequent build and regional incidents; one critical 2-hour Oregon service disruption (July 24) | Own report: about 8 hours on May 19–20, all workloads unreachable at peak | Comparable 2026 incident sample not established | Comparable 2026 incident sample not established |
| Company direction | Render Services, Inc.; flat workspace plans since April 2026 | Railway Corporation; $100M funding announced January 2026 | Kinsta Inc.; Sevalla continues Kinsta application hosting | Fly.io, Inc.; active app and managed database updates |
These selected tools have matching answers in this group. Turn off “Show differences only” to see them.
Sources for recovery, data and continuity
Sevalla app pricing, database pricing, deployments, terms and privacy, checked October 30, 2026. Example includes 100 build minutes and 1 GB external traffic. Fly.io pricing, Managed Postgres, cost management, changelog and terms, checked October 30, 2026. Example assumes one always-on Machine, 1 GB used database storage and 1 GB North America/Europe outbound traffic, before tax. Render pricing, disks, rollbacks, status history, terms; Railway volumes, volume limits, backups, point-in-time recovery, support answer, rollback, variables, incident report, funding report; Heroku filesystem, Postgres plans, releases, configuration, database export, February update, October update; Koyeb volumes, pricing, announcement, and deployment reference, secrets and domains, read September 7. Code recovery and data recovery are separate tasks.
What matters when choosing
Render is our Practical Pick for one person putting a small server-rendered Node app and a small Postgres database into production from Git. On Render’s free Hobby workspace, an always-on 512 MB web service costs $7 a month for 512 MB of memory and half a CPU, and the smallest paid Postgres database costs $6, or $13 a month together. Both have fixed monthly instance prices, prorated for time used. What separates Render is its documented recovery: paid Postgres includes point-in-time recovery (three days on Hobby), a paid disk is snapshotted daily, and a rollback redeploys a retained build with its settings. Its 2026 status history lists frequent build and regional incidents, but we found no outage that made every running service unreachable, as Railway’s did in May. The most important compromise is the plan boundary. Render describes Hobby as a plan for individuals’ personal projects and prototypes; HTTP request logs, longer log history and a second seat need Pro at $25 a month, and Render’s only spending limit covers build minutes. Render pricing, rollbacks, status history
If several people need access, or you would rather pay for measured usage, look at Railway. Pro is a $20 monthly minimum that includes $20 of usage, which a small app and database may stay within. Railway offers a hard spending limit on its paid plans, and a Railway support answer says Postgres point-in-time recovery is available on Pro. Weigh that against Railway’s own report of an outage of about eight hours on May 19 and 20, 2026 that, at its peak, left every Railway workload unreachable. Heroku remains reasonable for an app that already runs there. Its Basic dyno and smallest Postgres cost $12, but dyno files are temporary, its entry database has no point-in-time rollback and the platform has been in “sustaining engineering” since February 2026. Fly.io is worth considering for its managed database recovery and container deployment controls, but one small app and managed Postgres cost about $42.69 a month under our assumptions. Railway plans, Railway incident report and Heroku update (checked October 19, 2026); Fly.io pricing (checked October 30, 2026)
Render remains our recommendation, with a narrow lead. Prices and important configuration details for several other suitable platforms could not be confirmed.
This comparison covers managed platforms that build a long-running Node server from a Git repository and run it for you. It does not choose WordPress hosting, a static-site service, serverless functions or a self-managed virtual server. Where these platforms document rollback, it restores code and configuration, not database contents. Keep your code, configuration and secrets outside the platform, and keep your own database backups.
Our prices assume one always-on web service with about 512 MB of memory, the smallest paid managed Postgres, one person with dashboard access and normal monthly USD rates, before tax. Render, Railway, Heroku, Sevalla and Fly.io pricing was checked October 30, 2026. Fly.io’s example includes 1 GB of database storage and 1 GB of North American or European outbound traffic. They exclude bandwidth and build minutes beyond each plan’s allowance, extra storage and add-ons, and they are our arithmetic, not quotes. Railway bills by usage; at its published rates, a service averaging 0.5 GB of memory and a tenth of a CPU meters about $7 a month before its database (our illustration). Two conditions would change our recommendation: several people needing access, where Railway Pro’s $20 minimum compares with $38 for these services on Render Pro, or an app already on Heroku whose migration would cost more than its continuity risk. If you want Render’s request logs, or prefer to follow its Pro positioning for business use, budget $38; we would still choose Render for one person on its recovery and reliability record, but the price advantage moves to Railway. Before moving production data, restore a backup into a test database on the platform you choose.
Disclosure: Practical Picks articles are researched and written with AI, currently Anthropic’s Claude. Render and Railway each make an official Claude connector, Heroku offers Claude models, and Sevalla advertises Claude tooling. We have no relationship with these vendors or Anthropic over this article, and these AI features carried no weight in our judgment. Render connector and Railway connector (checked October 19, 2026); Heroku model pricing and Sevalla tooling (checked October 30, 2026)
Render
Preferred Pick · one small production app
Choose Render when one person runs a small Node app with a Postgres database and wants fixed instance prices, database point-in-time recovery and rollbacks to a retained build.
Reasons to choose it
- Fixed $7 instance and $6 database, no workspace fee on Hobby
- Postgres point-in-time recovery, three days on Hobby
- Rollback redeploys a retained build with its settings
What to watch
- Hobby is described for personal projects; a second seat or request logs means Pro at $25
- A disk blocks zero-downtime deploys and more than one instance
- Only build minutes have a spending limit
Render builds from a connected Git branch with native Node or a Dockerfile, using the build and start commands you set; the server must listen on the port Render provides. Its pricing page lists the 512 MB instance at $7 a month with 512 MB of memory and 0.5 CPU, billed by the second, and the smallest paid Postgres, with 256 MB of memory and 1 GB of storage, at $6. The Hobby workspace has no monthly fee and includes one seat, 25 services, 5 GB of bandwidth, 500 build minutes and two custom domains. Extra bandwidth costs $0.15 per GB. Web services, pricing
Recovery is the main reason we chose it. Paid Postgres includes point-in-time recovery for three days on Hobby and seven on Pro. A rollback redeploys a retained build with its start command and environment variables; Hobby keeps five builds and Pro 15. A rollback never changes a disk or database, and a rollback started from the dashboard switches off automatic deploys until you turn them back on. Pricing, rollbacks
Read the full Render analysis and sources
Choose another tool if a second person needs access. On April 23, 2026 Render replaced per-member plans with flat ones: Pro is $25 a month with unlimited seats, replacing Professional at $19 per member, and remaining workspaces moved over on August 1. Render describes Hobby as “for individuals building personal projects and prototypes” and Pro as for teams deploying production apps. Its terms, last modified July 10, 2026, contemplate commercial use, and we found no term that requires Pro for it. HTTP request logs, 14-day log history (seven on Hobby) and extra seats are Pro features, and Pro brings these services to $38 a month. Plan change, pricing, terms, logging
A persistent disk costs $0.25 per GB a month and attaches only to paid services. Render snapshots it every 24 hours and keeps snapshots for at least seven days, and a restore replaces the whole disk. A disk-backed service cannot run more than one instance, loses zero-downtime deploys, and its disk is unavailable to build, pre-deploy and one-off jobs. If you expect to scale, keep uploads in object storage or the database. Persistent disks
The free tier is for trying Render, not for production, and Render says so. A free web service spins down after 15 minutes without inbound traffic and takes about a minute to wake. A free Postgres database expires after 30 days, then is deleted after a 14-day grace period. Render’s only spending limit applies to build minutes: reaching it stops new builds, not the rest of the bill. Instance prices are fixed, so bandwidth, extra build minutes and storage are the variable charges. Free tier, billing FAQ
Render’s status history for May to October 2026 lists frequent incidents, most limited to builds and deploys or to one region. The only incident it marks critical for running web services was a two-hour disruption in Oregon on July 24, 2026. On October 14, builds and deploys did not start in any region for nearly three hours; that incident lists only builds and deploys as affected. Render offers contractual uptime commitments only on Enterprise. The contracting company is Render Services, Inc. Status history, pricing, terms
Used Render? Share your experience to help other readers choose.
Railway
Also consider · several people, usage billing
Choose Railway when several people need access and you prefer usage billing with a hard spending limit, after weighing its May 2026 platform-wide outage.
Reasons to choose it
- Pro includes $20 of usage and unlimited seats
- Email alerts and a hard spending limit
- Postgres point-in-time recovery over about four weeks
What to watch
- About eight hours on May 19–20, 2026, with all workloads unreachable at peak
- Hitting the hard limit makes workloads unreachable
- A service with a volume cannot use replicas and pauses on redeploy
Railway deploys from GitHub or a Dockerfile and documents an Express route, with generated and custom domains and automatic certificates. You pay for measured memory at $10 per GB a month, CPU at $20 per vCPU a month, outbound traffic at $0.05 per GB and volume storage at $0.15 per GB a month. Hobby is a $5 monthly minimum that includes $5 of usage; Pro is a $20 minimum that includes $20, with unlimited seats, 30-day logs and Railway support. A month with $7 of usage on Hobby costs $7, not $12. Plans, pricing page, Express guide, networking
Budget Pro for business use: Railway describes Hobby as for “hobbyist developers building side projects” and Pro as for teams shipping production applications, and an older Railway staff answer puts for-profit work on Pro. Its terms, effective April 20, 2026, do not spell out the boundary. The strongest reason to hesitate is reliability. Railway’s own report says Google Cloud wrongly suspended its account on May 19, 2026. The outage lasted about eight hours and, at its peak, left every Railway workload in every region unreachable. Its May 27 update acknowledged “reliability stumbles” and listed changes already made, including a distributed network router. Pricing page, staff answer, terms, incident report, summer update
Read the full Railway analysis and sources
Railway offers email alerts and a hard spending limit, with a $10 minimum. Reaching the hard limit makes your workloads unreachable, and Railway calls setting one “a possibly destructive action”. Per-replica resource limits set too low will crash a service. Railway’s own update describes a forgotten monitoring agent turning a $40 to $60 invoice into $544, which is why we suggest setting alerts on day one. Cost control, summer update
Railway’s optional sleep mode watches outbound traffic, not visitors. A service sleeps 5 to 10 minutes after its last outbound packet, so open database connections, telemetry and calls to other services keep it awake, and the first request after sleep may return an error while it starts. Serverless
Volumes mount when the container starts, not during build or pre-deploy steps. A service with a volume cannot use replicas and has a short outage on each redeploy. Scheduled volume backups keep daily copies for six days, weekly for 27 and monthly for 89. Postgres point-in-time recovery restores into a new service from a window of about four weeks, starting from the first backup after you enable it; the documentation does not state which plans have it, and a May 2026 Railway support answer says Pro. A rollback restores the previous image and custom variables; the plans page lists image retention of 72 hours on Hobby and 120 on Pro, and older deployments cannot be restored. Volumes, volume limits, backups, point-in-time recovery, support answer, deployment actions
Sealed variables cannot be read back through the usual command-line output and are not copied into duplicated environments, so keep your own record of secrets. Logs are kept for seven days on Hobby and 30 on Pro, and logging is capped at 500 lines per second per replica. The contracting company is Railway Corporation, which announced a $100 million funding round on January 22, 2026. Variables, logs, terms, funding report
Used Railway? Share your experience to help other readers choose.
Sevalla
Managed Git deployment with app and database pricing
Consider Sevalla for a managed Git workflow, while checking database recovery coverage before relying on it.
Reasons to choose it
- App and database managed by Kinsta
- Builds from Git with automatic deployment
- Rollback reuses an earlier image without rebuilding
What to watch
- Entry database restore window not established
- Rollback keeps current runtime settings and does not restore data
- Build minutes and external traffic add usage charges
Sevalla is Kinsta’s application-hosting platform. It builds from a Git repository and runs the app and database without asking you to manage the underlying servers. Choose it when you want that managed workflow and can accept less certain database recovery coverage than Render’s documented point-in-time recovery. Sevalla, about (checked October 30, 2026)
For this app, use the $10 S1 runtime with 1 GB RAM and 0.5 CPU, plus the $5 Database 1 with 1 GB storage. The $5 H1 runtime has only 0.3 GB RAM and no custom-domain support. At 100 build minutes and 1 GB external traffic, add $2 for builds and $0.10 for traffic: about $17.10 a month before tax. Internal connections are free. The example excludes optional persistent app storage; a larger database or more builds increases the bill. Application pricing, database pricing (checked October 30, 2026)
Read the full Sevalla analysis and sources
Its deployment documentation describes rollouts without downtime when the web process has a readiness probe. A rollback reuses an earlier image and its build-time variables, but keeps current runtime settings, processes and environment variables. It is not a database restore. Daily backups are advertised, but we could not establish the entry database’s restore window from the available documentation. Keep independent database exports. Deployments, database backup description (checked October 30, 2026)
Kinsta Inc.’s terms and privacy policy explicitly cover Sevalla. The terms allow termination through the platform at the end of the current subscription period. These documents identify the operator and its obligations; they do not establish uptime or successful recovery. We did not obtain a comparable incident history for Sevalla. Terms, privacy (checked October 30, 2026)
Used Sevalla? Share your experience to help other readers choose.
Fly.io
Managed database recovery and deployment control
Consider Fly.io when managed database recovery and container deployment controls justify more configuration and a higher starting bill.
Reasons to choose it
- Managed Postgres backups and high availability
- Point-in-time restore documented in August 2026
- Git-connected app deployment and portable container configuration
What to watch
- Managed Postgres starts at $38 before storage
- Overages have no ceiling and billing alerts are unavailable
- Major Postgres upgrades require creating a new cluster and copying data
Fly.io runs containers on small virtual machines and now offers Git-connected app deployment through its dashboard. Its August 2026 changelog describes repository reconnection and settings changes that open pull requests against your app configuration. Choose it when you want more control over deployment and placement and can take on more configuration than Render requires. Changelog (checked October 30, 2026)
The database makes this a more expensive starting point. The published 512 MB shared-CPU Machine costs $4.39 a month; Basic Managed Postgres costs $38, with database storage at $0.28 per GB-month. With 1 GB of database storage and 1 GB of North American or European outbound traffic at $0.02 per GB, that is about $42.69 a month before tax. This assumes one always-on Machine, shared IPv4 and no optional support subscription. Do not confuse the cheaper self-managed Postgres route with the managed database priced here. Pricing, Managed Postgres (checked October 30, 2026)
Read the full Fly.io analysis and sources
Fly.io says Managed Postgres includes automatic backups, high availability, failover and encryption. Its August 10 changelog documents point-in-time restore into a new cluster. Major database upgrades still require creating a new cluster and copying data with pg_dump and pg_restore; patch rollout follows maintenance windows without a fixed schedule. Those are documented capabilities and limits, not our own recovery results. Managed Postgres, changelog (checked October 30, 2026)
The main billing limit is the absence of an overage ceiling: Fly.io says it bills usage beyond allowances and does not yet provide billing alerts. Its terms call for written termination notice. Keep your own source, configuration and database exports, and budget for outbound traffic and database growth. An existing Heroku app may not justify the move just to gain these controls. Cost management, terms (checked October 30, 2026)
Used Fly.io? Share your experience to help other readers choose.
Which product are we missing?
Know a product that belongs in this article? Tell us what it does differently and why it deserves a closer look. Newcomers welcome.
Readers and product teams can suggest candidates. Suggestions inform our research; they do not guarantee inclusion or a recommendation.
Opens an email draft with this article attached as a link. Nothing is sent until you send it.
No email app? Send the product name, website, and this article’s title to admin@practicalpicks.guide.
What people are saying
Public user reports are a check on vendor claims, not a measured reliability score. An April 2026 Node community discussion includes support for managed platforms alongside arguments for running your own server; it frames the trade between management and cost, not a winner. On Trustpilot, read on September 7, 2026, Render reports include both billing complaints and praise for support, Railway’s profile includes downtime complaints and mixed feedback, and Koyeb reports include account-verification complaints. Self-selected reviews like these cannot support a numerical comparison, and we assign no reputation score.
Railway’s own summer update acknowledges reliability problems and billing surprises, and its May incident report is the most serious 2026 event we found among the original four platforms. Independent reporting on Heroku from February 2026 raised continuity questions; Heroku’s later releases have not settled them. GitHub’s 2022 incident account documents historical misuse of stolen Heroku integration tokens; it is not evidence of a current compromise.
Our October 19 search for newer complaints, incidents, ownership changes, lawsuits and regulatory actions involving the four companies found routine status incidents for Render and Heroku and tracker reports of Railway slowdowns on September 29 and 30 and October 3, 2026 that we could not confirm from Railway’s own records. None of it changed these conclusions. A search that finds nothing is not proof that nothing exists.
How we researched
This is public-source research. We did not create accounts, sign in, deploy an application, measure response times, restore a backup, contact support or test an export. No platform supplied access or payment for this work, and links carry no affiliate parameters. We first compared Render, Railway, Heroku and Koyeb on September 7, 2026 and rechecked the material claims on October 19, 2026: prices and plans, plan positioning, free-tier limits, spending controls, persistent storage, database recovery, rollback scope, log retention, company direction and 2026 incident records. Render’s pricing page loads its tables by script, so we read its server-delivered page text. Where we did not reread a page, the article says when we last read it.
We compare platforms for one person running a small server-rendered Node app with a Postgres database. Render is our recommendation; Railway suits several people who want usage billing. We judge fit, cost, recovery of code and data, reliability and company direction, and what it takes to leave. Sevalla offers a managed Git workflow; Fly.io offers more deployment control at a higher price.
On October 30, 2026 we checked additional public pricing, documentation and changelogs and examined DigitalOcean’s hosting alternatives. Sevalla joins this comparison for its managed Git workflow and app/database pricing. DigitalOcean App Platform, AWS Elastic Beanstalk, Azure App Service, Google Cloud Run, Google App Engine and Upsun also serve this kind of app; we could not establish complete comparable prices for them.
Heroku remains reasonable for an app already running there: Basic $7 plus Essential-0 $5 is $12 a month, but its entry database has no point-in-time rollback, dyno files are temporary and the platform remains in sustaining engineering. Its February announcement gives no end date, and October releases continued. Pricing (checked October 30, 2026); platform update, Postgres plans, October update (checked October 19, 2026).
Koyeb’s $29 Pro subscription, paid database charges and uncertain company direction make it hard to recommend for one small app. Its paid 512 MB eco-micro fits within the included $10 compute; Small Postgres at $0.04/hour plus $0.50 for 1 GB storage brings our 730-hour example to $58.70 a month. Its application volumes remain testing-only. Pricing, instances (checked October 30, 2026); volume limits (checked October 19, 2026).
No Rising Star selected. The Zeabur changelog shows public features before October 2024, and Leapcell’s blog dates its public beta to January 21, 2024. Sevalla continues Kinsta’s hosting product; a new brand does not make it a new entrant. Cloudflare Containers and Vercel Express come from established platform owners and use different routing or function models. Leapcell documents persistent servers as well as serverless hosting, but its database recovery and comparable pricing remain unclear. These sources were checked October 30, 2026; Sevalla’s about page, checked the same day, identifies its Kinsta origins. Several launch and independent review directories could not be read; this is not a claim that no suitable newcomer exists.
| Tool | Public sources | What remains uncertain |
|---|---|---|
| Render | Pricing page, April 2026 plan change, web service, free tier, disk, rollback, logging and billing documentation, July 2026 terms, 2026 status history. | Whether Render expects business apps on Hobby, real bills beyond allowances, restore success in practice. |
| Railway | Plans and pricing pages, cost control, sleep, volume, backup, point-in-time recovery and rollback documentation, April 2026 terms, May 2026 incident report and update. | Which plans have point-in-time recovery, a real usage bill, whether the May changes prevent a repeat. |
| Sevalla | App/database price tables, deployments, Kinsta terms/privacy, homepage and about page, checked October 30. | Entry database restore window, comparable incidents and actual usage bills. |
| Fly.io | Pricing, Managed Postgres, cost management, terms, privacy, security and 2026 changelog, checked October 30. | Comparable incident evidence, actual bills, database patch timing and recovery outcomes. |
| Newer entrants and other alternatives | October 30 hosting directory, Zeabur changelog, Leapcell blog and service documentation. | Several primary prices and launch directories could not be read or lacked required detail; no newcomer award follows from that uncertainty. |
None of this establishes deployment success, uptime, response times or security in practice. Before adopting a platform, rehearse a failed deploy, a database restore and a domain move with disposable data. Next substantive review is due December 18, 2026, or sooner if pricing, storage status or a platform’s direction changes. Found a changed price or term? Contact the editors.
Sources and evidence
Sources were checked October 19, 2026 unless another date is given. Pricing for Render, Railway, Heroku and Koyeb, Render billing and terms, Railway plans, cost control and terms, Heroku billing and Koyeb terms were checked again October 30; Fly.io and the alternative/newcomer sources marked above were checked October 30. Platform documentation establishes published offers, capabilities and terms, not observed performance. Status pages and incident reports are the platforms’ own accounts. Community posts and review-platform reports are attributed experiences, not representative samples.
- Render: pricing (server-delivered page text); April 23, 2026 plan change; web services; free tier; persistent disks; rollbacks; logging; billing FAQ; status history, May to October 2026; terms, July 10, 2026; privacy and security, read September 7.
- Railway: plans; pricing page; Express guide; public networking; cost control; serverless; volumes and volume limits; backups; point-in-time recovery and a May 23, 2026 support answer; deployment actions; variables; logs; incident report, May 19, 2026; summer update, May 27, 2026; staff answer on commercial use, about 2024; terms, effective April 20, 2026; privacy, read September 7; VentureBeat funding report, January 22, 2026.
- Heroku: an update on Heroku, February 6, 2026; October 2026 update; pricing; Node deployment; dyno filesystem; Postgres plans; Eco dyno hours; releases; logging; config vars; Postgres import and export; usage and billing; security policy and Salesforce privacy, read September 7.
- Koyeb: pricing; Mistral AI announcement, February 17, 2026; volumes, updated May 27, 2026. Read September 7: build from Git; scale-to-zero; secrets; domains; deployments; master services agreement, June 28, 2024; data-processing agreement.
- Community and independent sources: Reddit Node discussion, April 2026; Trustpilot Render, Railway and Koyeb, read September 7 (these pages did not load for our automated check on October 19); InfoWorld, February 9, 2026; GitHub security alert, April 2022.
This is a researched comparison, not a product review. Method: evidence standard, editorial picks.