Hey HN, I’m Cyw, the founder of Agentainer (https://agentainer.io/), a platform designed to deploy and manage long-running AI agents with zero DevOps. We just launched the first open source version of Agentainer: Agentainer Lab (https://github.com/oso95/Agentainer-lab) on GitHub.

Little bit of background: most infrastructure today is built for short-lived, stateless workloads—Lambda, Cloud Run, or even Kubernetes pods. But AI agents aren’t like that. They’re long-running processes with memory, history, and evolving state. Running them reliably in production usually means gluing together a bunch of services (volume mounts, retry queues, crash recovery, gateways, etc.) just to approximate what a simple web app gets out of the box.

To make my life easier when deploying agents for projects (both personal and work-related), I started designing an infrastructure layer that could treat agents as durable services from day one. No YAML. No juggling services. Just give it a Docker image or Dockerfile, and Agentainer handles the rest. Basically, a Vercel-like solution.

Each agent runs in its own isolated container, with persistent volume mounts, crash recovery, and queued request replay. If an agent crashes mid-task, it restarts and picks up where it left off. Agentainer gives every agent a clean proxy endpoint by default, so you don’t have to worry about port management or network config. Oh, if you’ve ever built long-running agents, you know how important checkpoints are—I got it taken care of already. (Check out: https://github.com/oso95/Agentainer-lab/blob/main/docs/RESIL...)

Everything is CLI-first and API-accessible. In fact, I originally built this so my own coding agent could manage infrastructure without burning tokens repeating shell commands lol. You can deploy, restart, or remove agents programmatically—and the same flow works in dev and prod.

I did some math, and for the right workloads like agentic backends with frequent requests or persistent state, this architecture could reduce cloud costs significantly, even by 30~40%, by replacing per-request billing and minimizing infra sprawl. We’re still early, but excited to see what others build on top of it.

Anyway, right now Agentainer Lab is focused on local dev and self-hosting. The bigger Agentainer.io roadmap includes observability, audit logs, backup/restore, and full auto-scaling to unlock the full experience. If you’re interested, you can sign up for early access on our website, we’ll only send you one email when the production version launches, and then your email will be deleted from our database.

GitHub: https://github.com/oso95/Agentainer-lab Platform: https://agentainer.io

Would love to hear feedback from others working on LLM agents or trying to run stateful workloads in production. What’s your current setup? Do you think this can help you?