Run Agones yourself if your studio already operates Kubernetes in production and wants to own every layer of its game server stack. For most other studios a managed game server host costs less and ships sooner, because the real cost of Agones is the clusters and the people who run them, not the software.
Agones is a mature, well-documented open source project. This post covers what running it involves, what it costs in total, how the managed options compare, and how to move from Agones to Gameye.
The options at a glance
| Self-hosted Agones | Gameye | AWS GameLift | Edgegap | |
|---|---|---|---|---|
| What you operate | Kubernetes clusters, Agones, Fleets and autoscalers in every region | A Docker image and one API call | GameLift fleets on AWS | Containers on Edgegap’s network |
| SDK in your game server | Agones SDK required | None | GameLift Server SDK required | Edgegap SDK available, optional |
| Compute price | Free software. You pay for your own nodes, which run around the clock | $0.07/vCPU-hr on demand, $0.035 reserved, billed per second | About $0.044 to $0.063/vCPU-hr for gen-6 and gen-7 Linux on demand | About $0.069/vCPU-hr plus about $0.0051/GB-hr of RAM |
| Bandwidth | Your cloud’s egress, about $0.09 to $0.12/GB on AWS, GCP and Azure | Included | Free on gen-6 and newer. Gen-5 and older, and China, about $0.09/GB | $0.10/GB |
| Where it runs | Anywhere Kubernetes runs, including your own hardware | 21 providers and 200+ data centers, bare metal first | AWS only | 615+ locations across 17+ providers |
| Hardware | Whatever you provision | Dedicated bare metal, 4.7 GHz fleet average | About 2.8 to 3.9 GHz virtual cores | Public pool varies from 2.4 to 3.2 GHz |
What running Agones really involves
The Agones docs describe it as “a library for hosting, running and scaling dedicated game servers on Kubernetes.” It adds GameServer, Fleet and FleetAutoscaler resources to Kubernetes. Your team builds and runs everything around those resources.
- Clusters. You provision a Kubernetes cluster in each region you serve, install Agones on it and size the node pools. Agones runs in any cloud or on your own hardware.
- Game server integration. The Agones docs say the client SDK is required for a game server to work with Agones. Your server calls
Readywhen it can take players and sends regularHealthpings. If the pings stop, Agones marks the server Unhealthy. - Fleets and autoscaling. A Fleet keeps a pool of GameServers warm. A FleetAutoscaler resizes it on a fixed interval. The Buffer policy keeps a set number of Ready servers, and Webhook, Counter and List policies cover custom logic. You tune these per game mode and per region.
- Allocation and routing. Your matchmaker requests a server through a GameServerAllocation. Routing players across regions and clusters is code you write and maintain.
- Upgrades. The Agones docs recommend upgrading by creating a new cluster, sending a small share of matches to it, and raising that share until the old cluster can be shut down. On a single cluster they recommend a maintenance window with the game offline. They also recommend doing Kubernetes minor upgrades together with the matching Agones upgrade. In-place upgrades are in Beta, and the docs say to use the multi-cluster strategy for production-critical upgrades.
- SDK versions. A server built on Stable SDK APIs is guaranteed compatible for 10 Agones releases, about a year. Plan SDK updates into your build schedule.
- On-call. When a node or a controller fails during a match, your team gets paged, not a vendor.
This is the normal work of running your own platform, not a flaw in Agones.
Total cost is infrastructure plus engineering time
Agones has no license fee. The bill comes from three places.
- Nodes. Cluster nodes run whether matches are active or not, and you size each region for its peak.
- Bandwidth. AWS, GCP and Azure charge about $0.09 to $0.12 per GB of outbound traffic. That line grows with every player.
- People. Someone owns cluster upgrades, security patches, scaling rules, multi-region routing and on-call.
The Gameye vs Agones comparison models a mid-size deployment of about 5,000 concurrent players across three regions:
| Cost line | Self-hosted Agones |
|---|---|
| Node compute | About $8,000 to $15,000 a month |
| Egress | About $3,000 to $8,000 a month |
| One DevOps engineer | $150,000 to $200,000 a year |
| Total | About $25,000 to $40,000 a month |
The same load on Gameye comes to roughly $8,000 to $12,000 a month, with bandwidth included. Treat both figures as a model, not a quote. Your numbers depend on CPU use per player, traffic per player and how much your player count swings during the day. The cost calculator lets you enter your own load profile for Gameye, Edgegap and AWS.
Utilization matters too. A self-run cluster costs the same at 4 a.m. as at peak unless you scale nodes down, and new nodes take minutes to start when players return. Gameye bills on-demand capacity per second of container runtime, and a new container starts in 0.5 seconds on average.
When self-hosting Agones is the right call
Self-hosting is the better choice in these cases.
- You already run Kubernetes in production. If a platform team already runs clusters for your backend, Agones is incremental work, and your game servers share the same tooling.
- You need servers on your own hardware. Agones runs anywhere Kubernetes runs, including on-premises. A managed host runs on its own providers.
- Your scheduling is unusual. Webhook autoscalers let you plug your own scaling logic into a Fleet, and you can change the open source code itself.
- Your load is large and steady. If your player count keeps your own nodes busy around the clock, idle capacity costs little and the engineering cost is spread over more players.
If none of these apply, the engineering time usually costs more than a managed host.
How the managed options differ
All three run your servers for you. They differ on the SDK, bandwidth, hardware and where they run.
- AWS GameLift suits studios already deep in AWS. It includes FlexMatch matchmaking and spot instances through FleetIQ. Outbound bandwidth has been free on gen-6 and newer instances since June 2026. Your server needs the GameLift Server SDK, and fleets run on AWS only. See Gameye vs AWS GameLift.
- Edgegap suits games whose players are spread thinly over many regions. It offers 615+ locations, a free tier and self-serve onboarding. Egress costs $0.10/GB on top of compute, and hardware in the public pool varies from 2.4 to 3.2 GHz. Its recommended placement strategy needs your matchmaker to send player IP addresses to its API. See Gameye vs Edgegap.
- Gameye runs your existing container with no SDK. It places every session on bare metal first and only uses cloud when bare metal in that region is full. The fleet averages 4.7 GHz, on AMD Zen 4 and Zen 5 where available. Bandwidth is included. Gameye offers a 99.7% contractual uptime SLA, and measured uptime over the last 12 months is 99.9%.
Migrating from Agones to Gameye
Your game server already runs in a container for Agones, so most of the work is removing code.
- Push the same image. Gameye runs the Docker image you already build.
- Remove the Agones SDK calls. Take out
Ready,Health,Shutdownand any other Agones SDK calls. Gameye needs no replacement SDK. - Replace the allocation call. Where your matchmaker created a GameServerAllocation, it calls
POST /sessionwith a region and your image. Gameye returns an IP address and port. - End sessions. Call stop when the match ends, or set a TTL and let Gameye end the session.
- Shift traffic in steps. Send a small share of matches to Gameye, compare tick times and cost, then raise the share. This is the same pattern Agones recommends for its own upgrades.
Your Fleets, FleetAutoscalers and cluster configuration have no Gameye equivalent to port. Placement and scaling happen behind the API. See how it works for the session flow and pricing for the full rate card.
Summary
- Self-host Agones if you already run Kubernetes in production, need your own hardware or custom scheduling, and will staff upgrades and on-call for the life of the game.
- Count the full cost of Agones. That means nodes that run all day, cloud egress and at least one engineer, not just the free license.
- Use a managed host if you would rather spend engineering time on the game. Compare SDK requirements, bandwidth pricing and hardware.
- Moving from Agones to Gameye keeps your container, removes the Agones SDK calls and replaces your allocation logic with one API call.
To test it with your own server, start a free trial and run a match on Gameye, or talk to an engineer about moving off your Agones clusters.