ttl64 is in beta

You know the cluster. The one somebody spun up "just for a quick test" three months ago. Nobody remembers what's running on it, nobody wants to be the one who deletes it, and it shows up on the bill every month like a gym membership.

We've all had one. ttl64 exists so you don't have to.

The idea is simple. Every cluster you create has a TTL, a time to live. When it runs out, the cluster is deleted. No reminder emails, no cleanup script, no guilt. You try the Helm chart, reproduce the bug, test the upgrade, and then you walk away.

Today we're opening the beta.

What you actually get

A real K3s cluster. Each node is its own microVM with its own vCPU and memory, a control plane plus however many workers you ask for. It isn't a namespace on somebody else's cluster, and it isn't a simulator pretending to be one.

From create to a working kubeconfig takes about 15 seconds. That's fast enough that waiting for a cluster stops being a reason not to make one.

Once it's up, it behaves like any other cluster. kubectl, Helm and your CI don't need to know it's temporary. A LoadBalancer Service gets a real name under lb.ttl64.computer, and if you expose it as HTTP from the dashboard you get TLS too. You can add volumes when a workload needs to keep data around while the cluster lives. And if you'd rather not install the same things every time, there are templates with Argo CD, kube-prometheus-stack or PostgreSQL already in place.

A TTL can be anything from 30 minutes to 24 hours. Need more time? Extend it. Done early? Delete it, and billing stops at the end of the hour you started.

Try it

$ curl -fsSL https://ttl64.computer/install | sh
$ ttl64 login
$ ttl64 create helm-test --workers 1 --ttl 1h

The CLI stays with you while the cluster comes up and tells you what it's doing. After that, ttl64 kubeconfig helm-test hands you the kubeconfig. If you're on a network where you'd rather not pass credentials around, ttl64 ssh helm-test drops you into a shell inside the cluster with kubectl and helm ready to go.

If you prefer to look at things, the web app draws your cluster as a live canvas: services, load balancers and volumes, exactly as Kubernetes reports them, moving as they change. It's oddly satisfying to watch.

Your AI can drive it

There's an MCP server at mcp.ttl64.computer, so Claude, Codex or Grok can create a cluster, run kubectl in it and clean up after themselves. Handing an AI the keys to your infrastructure should make anyone a little nervous, so each one gets its own key: read-only or full access, a monthly spending cap and an expiry date. The kubeconfig never reaches the model. The AI page has the details.

What it costs

You pay for computers while they exist, by the started hour. A control plane and one worker means two computers.

Size	Per computer	Price per hour
Small	1 vCPU · 2 GB	R$ 0,69
Medium	2 vCPU · 4 GB	R$ 1,29
Large	4 vCPU · 8 GB	R$ 2,39

Plans come with computer-hours included every month.

Developer is R$ 49 a month. It gives you 50 computer-hours on Small computers, up to 2 computers per cluster, 2 clusters at once, TTLs up to 8 hours and one region.

Pro is R$ 129 a month. It gives you 160 computer-hours on any size, up to 4 computers per cluster, 5 clusters at once, TTLs up to 24 hours, every region and early access to whatever we're building next.

Load balancers are R$ 30 a month and storage is R$ 1 per GB-month, both charged by the hour while they exist. You can pay by card or Pix.

Where it runs

Our first region is Boca Raton, in the United States. São Paulo is next, and we can't wait to have clusters running close to home. ttl64 regions will tell you what your plan can use.

About the word "beta"

We mean it. Things will change, and some things will break, probably at the worst possible moment. When that happens, please tell us. Open a ticket from the dashboard, run ttl64 support "what went wrong" --cluster NAME, or just email us. Tickets opened about a cluster arrive with its context attached, so you don't have to explain where it runs or what it looks like. A real person reads every one of them.

Thank you for trying something this early. It genuinely means a lot to us.

Create your first cluster, and let it go when you're done.