Railway's developer experience on servers you own.

Why it costs less

The workload: 3 web services, Postgres and Redis, 4 vCPU and 16 GB RAM in total, 100 GB egress and 50 GB of volume storage per month, one production environment, 730 hours, no support plan, no tax.

List prices read on 2026-10-22; check current prices. Exchange rate: 1.15 USD per EUR (an assumption; check the rate on the day you compare). 5050cloud itself costs $0. Railway list prices used: $20 per vCPU month, $10 per GB of RAM month, $0.05 per GB egress, $0.15 per GB of volume month (source, 2026-10-22).

• Low average use favours Railway. At typical load (an average of 1 vCPU and 8 GB) Railway costs about $112.5, about the same as a Hetzner CCX23 and less than a DigitalOcean, AWS on-demand or Azure VM. For idle hobby projects there is no saving to claim.
• Egress is included on Hetzner in the EU (20 TB) and on DigitalOcean (5 to 6 TB).
• Growth is flat on a VM. Twice the traffic costs nothing until the box is full; the next step is a fixed price you choose.
• You own patching. The installer turns on unattended security updates and 5050 upgrade reverts itself when the new version is unhealthy. Reboots are yours.

See your own number: the Usage page shows what your use would cost on Railway.

Headline: Hetzner CCX23 + 5050cloud: about $102 a month, 60% less than Railway at steady load.

Use every core you pay for

Processes set to auto gives a single-threaded Node service at least 1.6x the throughput of one process.

Most Node apps run one process and use one core. Turn on Use all cores (Settings > Scale) and 5050cloud runs one process per vCPU. For a start command of the form node server.js it runs the processes in Node cluster mode sharing one port; for every other runtime it sets WEB_CONCURRENCY, which gunicorn, uvicorn, puma and many frameworks read. Test condition: examples/node-cpu on a 2 vCPU non-burstable VM, load from a second host, 60 s warm-up, median of 3 runs of 60 s. The VM size is named in report.json. The test app and script are in the repo.

Use all cores guide

Add servers with one command

curl -fsSL https://5050cloud.com/install.sh | sudo sh -s -- --join https://console.example.com --token J-... --roles app

• A second host joins with one command and is online in under 2 minutes. Get the command from Hosts > Add host or 5050 host add.
• The agent dials out: no inbound management ports, works behind NAT.
• Replicas across hosts with built-in load balancing.
• Private networking over WireGuard.
• A second public entry point with --roles ingress.

Reliable by default

• Database backups to any S3-compatible bucket with point-in-time restore to a copy.
• Zero-downtime deploys for services without a volume, for public and private traffic.
• The old version keeps serving if a deploy fails.
• Rollbacks restore the variables a deployment ran with and do not expire after a few days.
• 5050 upgrade reverts itself if the new version is unhealthy.
• Apps keep serving while the 5050d process is stopped or restarting, and across a worker reboot.
• The control plane's state streams to your bucket and rebuilds on a new VM.

v0.1.0 is an early release. A fresh single-host install was smoke tested end to end (install, sign-up, image deploy, generated HTTPS domain, Postgres with a variable reference). The full release suite on fresh VMs, which proves each line above, runs next. The test conditions are listed in claims.json.

Move from Railway in minutes

• 5050 import railway brings services, variables, domains, volumes and databases over as staged changes.
• railway.json in your repo is honoured.
• The railway CLI alias, the same variable names and the same reference syntax: ${{Postgres.DATABASE_URL}} works as it does on Railway.

Coming from Railway lists every difference.

No lock-in

• Builds use Railpack, Railway's own builder, or your Dockerfile.
• Every deployment is a standard Docker image.
• 5050 config pull exports an environment's services and settings to one file.
• The server is AGPL-3.0. The CLI, API types, templates and installer are MIT.

Compared with

Facts as of 2026-10-22, from each project's docs. Details per tool under Compare.

Honest limits

• v0.1.0 is an early release. A fresh single-host install on Ubuntu 24.04 was smoke tested: sign-up, an image deploy, a generated HTTPS domain and Postgres with a variable reference. Multi-host, backups, upgrades and the Railway import are built but have not yet been through the full release test suite.
• One control host. With one host, a host failure stops everything until it is rebuilt. With more hosts, apps on other hosts keep serving (with a second ingress host), but deploys, cron and the console stop until the control host is back or rebuilt.
• Cron runs pause while the control plane is down; missed runs are listed, not replayed.
• Databases run on one host each. No high availability, no automatic failover.
• Services with a volume have a short gap on every deploy.
• Without a second ingress host, all public traffic enters through one machine. No global edge, CDN or DDoS protection.
• No autoscaling. Volumes have no size cap; they share the host disk.
• With a bucket, control plane state is replicated within seconds; without one, the recovery point is the last hourly snapshot on the same disk.
• Metrics are kept 30 days. Logs are kept 7 days on each host.
• IPv4 only. Ubuntu 24.04 is the tested OS.
• You own the hardware and the OS (security updates are automated, reboots are yours).

Details for each limit

FAQ