How do I set a hard usage limit on Railway?

Configuring your usage limits in Railway

To change these settings in the web UI, you must hold the workspace admin role. If you run multiple accounts or team setups, switch to the right target first by opening the account switcher in the top left corner of the Railway dashboard and clicking into your workspace.

Once you are in the workspace:

1. Go to the Workspace Usage page.
2. Click the Set Usage Limits button.
3. In the modal that opens, look at the separate sections for Compute Usage and Agent Usage.
4. Set a Custom email alert amount if you want an early warning. This acts as a soft limit; Railway emails you when your billing cycle spend reaches that dollar mark, but your containers keep running without interruption.
5. Set your Hard limit amount. When total usage crosses this figure in your active billing period, Railway takes your workloads offline to halt billing.
6. Save your changes.

If you spend your day in a terminal, you can manage these limits directly through the Railway CLI. To see what you currently have configured across your workspace, run:

railway usage limit status

To update the compute spending caps for your entire workspace, pass your soft warning threshold and your hard shutdown threshold:

railway usage limit set --target workspace --soft 75 --hard 125

Agent spending uses its own CLI target. To cap LLM consumption separately, run:

railway usage limit set --target agent --soft 7.50 --hard 20

Both targets accept separate soft alerts and hard caps, giving you room to handle alerts before an automated shutdown triggers.

Compute limits vs Agent limits: how they behave

Railway does not pool your infrastructure spend and your AI agent spend into a single bucket. Compute Usage and Agent Usage are tracked and limited independently, and hitting a hard ceiling on one produces a completely different outcome than hitting a ceiling on the other.

Compute Usage covers your core container resources: CPU, memory, storage, and network egress across all services running in the workspace. If you hit your Compute hard limit, Railway takes all of your service workloads offline immediately, but the Railway Agent remains available. You can still use the agent to inspect configurations, troubleshoot your project, or adjust your settings.

Agent Usage covers Railway Agent LLM consumption. By default, Agent Usage has a hard limit of $5 on the Hobby plan and $20 on the Pro plan. You can raise or lower this agent limit at any time on the Workspace Usage page, but unlike the Compute limit, the Agent limit cannot be removed entirely. If your agent hits its hard ceiling, Railway disables the agent, but leaves your running services and container workloads completely untouched.

The base subscription plans determine your default resource ceilings, included credits, and baseline spending before usage overages kick in:

Included usage resets at the end of every billing cycle and does not roll over. If your usage stays within your included balance, you only pay the base plan subscription fee. If your usage exceeds the included credit, you pay the difference on top of your plan price.

What happens when you hit the hard limit?

When you configure a hard limit on Compute, Railway sends three reminder notifications to make sure you are not caught off guard by an unexpected outage:

• An email when your billing cycle spend reaches 75% of your hard limit.
• A second email when your spend reaches 90% of your hard limit.
• A final email when usage reaches 100% of your limit, confirming that all workloads have been taken offline.

Workloads shut down fast. Taking workloads offline stops compute and memory billing immediately, which protects you from an unchecked bill, but it creates a hard production outage until you manually raise the limit or a new billing cycle resets your spend.

If you use credits as your payment method instead of a standard card, watch your balance closely. When an account using credits hits a zero balance, Railway cancels your subscription, disables all deployments, and stops every running workload in the account. For trial users, new accounts receive a one-time grant of $5 in free credit, and you cannot purchase additional credits without upgrading to the Hobby plan.

Fine-tuning costs without taking services offline

Hard limits are an emergency backstop, not a scaling strategy. If you want to keep costs predictable without dropping traffic, Railway provides three other built-in cost controls: resource limits, private networking, and serverless auto-sleeping.

Service replica limits

By default, container RAM costs $10 / GB / month ($0.000231 / GB / minute) and container CPU costs $20 / vCPU / month ($0.000463 / vCPU / minute). If a service runs runaway loops or memory leaks, it can scale up to the plan ceiling. To restrict this, navigate to your service settings > Deploy > Replica Limits. You can set the maximum CPU and memory each replica can consume. Be careful with this setting: setting replica limits too low will cause your service to crash if it exceeds the limit.

Route traffic over private networking

Network Egress costs $0.05 / GB ($0.000000047683716 / KB). If you run an application container that talks to a database inside the same project, routing queries over public hostnames charges egress fees on every single query. You can avoid network egress costs completely by using Private Networking for internal service communication. In your database connection configuration, point your application to the DATABASE_URL environment variable rather than DATABASE_PUBLIC_URL.

Turn on serverless auto-sleeping

For staging environments, internal tools, or low-traffic services, running containers 24/7 wastes money. You can enable Serverless to stop a service when it is inactive, which cuts the ongoing compute bill down to zero while idle. To enable this, go to your service settings > Deploy > Serverless, and toggle "Enable Serverless". Railway brings the service back up as soon as a new request arrives.

Monitoring spend and setting brakes with Costbrake

Railway's native hard limit is workspace-wide: when you hit it, every service goes down. If you need earlier visibility, granular alerts, or want to halt a single broken service without taking your entire database and API offline, you can pair Railway with Costbrake.

Costbrake monitors your spending across 23 cloud, AI and SaaS providers, checking your usage hourly on the Free plan and every 5 minutes on Pro and Team plans. If spend in an hour jumps higher than 3x your 7-day baseline and runs at least $5 above normal, you get a spike alert over Slack, Discord, email, webhooks, or PagerDuty. You also get forecast alerts when the last 72 hours of spending puts you on track to bust your monthly budget, and budget warnings at 50%, 80%, and 100%.

For Railway, Costbrake reads workspace usage every 5 minutes using a standard workspace API token. When something goes wrong, you can use two opt-in brakes from the Costbrake dashboard:

• Hard limit at current usage: Sets the workspace hard limit directly to your current usage rounded down to whole dollars (Railway's minimum limit is $10). Railway will immediately take workloads offline until you raise it. An undo button restores the old limit, and Railway redeploys your stopped services.
• Stop service: Stops the running deployment of one service in one environment. Nothing gets deleted; you stop the bleed on a single misconfigured worker while keeping your database and web frontend live.

Costbrake never triggers a brake automatically; an admin must click and confirm. You can review plan options on the Costbrake pricing page or follow the agent setup walkthrough to connect your workspace. For broader context on how platforms handle spending caps, check out our guides on How does Railway pricing work? and Which cloud and AI providers let you set a hard spending cap?.