Eveland: run 100 agents on your own server

[mechiland]

tldr;

Opensource self-hosted eve agents runtime: https://github.com/evelandhq/eveland

---

and here is the story:

Hey,

I was really excited when seeing eve agent for the first time - 3 month ago. Exactly the same excitement when I see Next.js 8 years ago. As someone trying to deploy agents inside company after using OpenClaw, Hermes, mastra, pandatic...I believe eve is the perfect shape to develop and deploy agents in a company.

However, there is one issue - do I really have to deploy all my agents to Vercel? I quickly found that we are running dozens of agents inside company, and most of the agents accessing self-hosted services, like finance backend or services inside cooperate network.

So...I said - why not self hosted? the story begins. and here is what you see:

• catchup with latest version of eve (an interesting journey with a bit pain)
• using bwrap sandbox instead of docker to maximize server ultilization
• re-write a shared workflow so it's fully multi-tanent support running multiple agents
• Support Otel so that you can trace all aspect of your agent

Some screenshot:

Screen	Page
Home
Playground
Sessions
Logs
Schedules
Usage
Source
Instance Health

We have use eveland hosting ~20 agents on a 4core 16GB machine internally for over 2 month, and we enjoy it. We couldn't believe that the "old" days - leaving laptop lid opened in office running hermes/openclaw and worrying about the next version will break... Now with eveland + eve agents, we have a stable, tracable, controllable infrastructure. Vercel is great. And eveland is to help.

Hope it help you as well.

[lucamaraschi]

And if you want to run Eve at scale in K8s...here is your solution https://blog.platformatic.dev/run-durable-eve-agents-on-kubernetes-with-platformatic

[mechiland]

Thanks for sharing, @lucamaraschi — nice write-up, and it's good to see more self-hosted paths for Eve.

I think the two projects are complementary rather than competing; they answer different questions:

• Platformatic answers "how do I run one Eve agent as a normal K8s workload, with multiple pods and version-pinned durable runs". If a team already operates Kubernetes and needs an agent to scale horizontally, that's a natural fit.
• Eveland answers "how do I run dozens of low-traffic agents on one box, cheaply, and actually see what they did". It's a whole platform layer on top of Eve: import from Git, immutable releases, a shared gateway with per-project URLs and login, secrets, platform-owned cron, session/usage tracing. Each agent runs as a hardened systemd unit and scales to zero after a few minutes idle, which is why ~20 agents fit on a 4c/16G machine. No Kubernetes involved, by design.

The workflow story is closer than it looks: both replace Vercel World with a Postgres-backed one. Ours (@evelandhq/workflow-world) is multi-tenant in a single database (tenant_id on every row, partitioned per project) with an external dispatcher that is the only thing allowed to claim jobs — we learned the hard way that in-process runners on a shared queue end up claiming other projects' work. Runs are pinned to a deployment id too, so a session finishes on the code it started on.

One thing the post didn't cover that I'm curious about: which sandbox backend do you use inside the pod? Without a Docker daemon or KVM, Eve's default chain falls through to just-bash, which can't run real binaries. We ended up shipping a bubblewrap backend for that reason. Would be interested to hear how you handle it.

The version-draining logic in ICC (checking active runs, pending hooks, waits, queued messages before retiring a version) is a nice touch — that's something I'd like to add on our side.