Usage limits and best practices
What your plan includes for application hosting, what actually consumes it, and how to get more out of it before you need a bigger one.
This page covers application hosting — the apps, agents, services and data you deploy. Dedicated model serving is a separate product with its own capacity model; see Nexlayer Inference.
Paid application plans are flat, not metered
There is no per-request, per-build or per-token counter to watch, and no overage invoice. What your plan gives you is a capacity envelope — a reasonable amount of running infrastructure for one team's work. Almost nobody reaches the edge of it. If your usage sits consistently beyond what the plan is sized for, we contact you directly and size the right plan with you. Your apps are not throttled or shut off without that conversation.
What consumes your plan
Running applications
Your apps stay on. They do not sleep between requests, so a deployed app occupies capacity whether or not it is serving traffic right now.
Services inside a project
A frontend, an API, a database, a cache and a vector store deploy as one application — but each one runs. A five-service project uses more than a one-service project.
Autoscaled replicas
Apps scale out on CPU and memory pressure. A workload that sits near its thresholds runs more replicas, for longer.
Persistent volumes
Storage attached to databases and stateful services persists across deploys and counts while it exists — including on projects you have stopped using.
Builds
Every deploy builds and pushes an image. Frequent pushes to the same project cost more build capacity than batching them. On Free, builds may queue behind other work; Pro builds do not.
Backups and snapshots
Point-in-time snapshots consume storage, and retention scales with your plan.
What doesn't
Bandwidth
Zero egress fees. Serving traffic out of your app is not billed and does not count against your plan.
Agent deploys
The Nexlayer agent building and deploying your app is free on every plan, including Free.
Projects
Unlimited application projects on Pro. The count of projects is not the limit — what those projects are running is.
Inference
Model serving is a separate product with its own capacity model. It is not drawn from your application plan.
Getting more out of your plan
In rough order of how much difference they make.
Delete the projects you are done with
This is the single biggest lever, and the one most people miss. An experiment from three months ago is still running, still holding its volume, and still occupying capacity. Nexlayer keeps apps on by design — that is the product — which means nothing stops on its own.
Batch deploys instead of pushing every commit
Each push triggers a build. While you are iterating, work locally and deploy when the change is worth a build. Your agent will happily deploy forty times an hour if you let it.
Right-size services before you scale them
A service pinned near its CPU or memory ceiling autoscales constantly. Raising its limits once is usually cheaper than running extra replicas permanently. See Scaling.
Use one project per application, not per service
Frontend, API, database and cache belong in one nexlayer.yaml. Splitting them across projects gives you the same running services plus the overhead of several deployments to keep in sync.
Prune volumes and snapshots you are not restoring from
Storage is the quietest consumer on the list, because nothing about a full disk shows up until something fails to write.
Keep staging small
A staging copy that mirrors production doubles what you run. Staging rarely needs production's replica counts or volume sizes.
Where your plan sits
Free
One workspace, a full-stack application that stays on, real persistence, and agent deploys. Builds may queue behind other work.
Pro
One personal workspace, unlimited application projects, persistent volumes, custom domains and unqueued builds — at one flat monthly price with no metering.
Enterprise
Unlimited workspaces, projects and teams, multi-region and reserved capacity, SSO, RBAC and audit logging, with SLAs and 24/7 incident response.
If you outgrow your plan
We reach out before anything changes. Most conversations end in one of two places: Enterprise, if what grew is the application side — more teams, more workspaces, reserved capacity, compliance requirements — or Nexlayer Inference, if what grew is model serving, which is bought as dedicated capacity rather than as a hosting tier. You can also start that conversation yourself.