Neckrez

Servers for AI agents, rented by API

Give an AI agent its own disposable servers: a REST API with revocable keys, an llms.txt it can read in one request, hourly billing and a prepaid crypto balance as the spending limit.

In short

An agent that writes and runs code sooner or later needs a machine to run it on: a sandbox for scripts it does not trust, a worker for a long job, a browser for research, a box for a tool it built itself. Your own laptop is the wrong place for that. A disposable server with a hard budget is the right one.

Neckrez has a plain REST API for this. A person signs up with an email, tops up a balance in crypto and creates an API key; from then on the agent can deploy, poll, stop and delete servers on its own. We also publish /llms.txt, the whole API in one plain-text file that a model reads in a single request.

What an agent can do

  • Deploy a server with one POST: name, plan, city and a root password. It is running in about a minute and the response carries its IPv4.
  • Deploy several at once. The bulk call takes up to 10 servers per request and reports which ones were created and which failed.
  • Stop, start, reboot and delete. Billing stops the moment a server is deleted.
  • Take a snapshot of a server and restore it as a new one. Snapshots are kept 7 days and the restored server gets a new IP.
  • Read its own balance before spending, so it can decide not to deploy instead of failing with an error.

The first account and the first key are made by a person: the key is created in the dashboard, shown once, and starts with nek_. After that nothing needs a human, apart from topping up the balance.

A deploy in three calls

Replace the key, the password and the id with your own. The full list of endpoints, fields and error codes is in the API docs and in /llms.txt.

# deploy
curl -s https://neckrez.tech/api/v1/servers \
  -H "Authorization: Bearer nek_..." \
  -H "Content-Type: application/json" \
  -d '{"name":"sandbox-1","plan_id":"quark","location_code":"AMS","root_password":"<a strong password>"}'

# poll until "status" is "running", then use "ip"
curl -s https://neckrez.tech/api/v1/servers/<id> -H "Authorization: Bearer nek_..."

# destroy: hourly billing stops
curl -s -X DELETE https://neckrez.tech/api/v1/servers/<id> -H "Authorization: Bearer nek_..."

A deploy is quick but not instant, so an agent should poll the server about every 5 seconds until the status is running rather than assume it is up. Money values come back as strings: parse them as decimals, not floats.

The balance is the leash

The most useful limit on an agent is money it cannot exceed. Neckrez is prepaid: you fund the balance with exactly what you are willing to spend, and every running server draws from it by the hour.

  • A deploy is refused without funds. The balance has to cover at least the first hour, otherwise the API answers 402 and nothing is created.
  • An empty balance stops the spending. When the balance can no longer cover a server it is archived: the disk is snapshotted, the server is deleted and the snapshot is kept for 7 days. Bare metal cannot be snapshotted and is deleted with its data.
  • There is no invoice at the end of the month. What you topped up is the most the account can ever cost.
  • We email you when the balance runs low, and the billing page shows how many days it lasts at the current rate.

Be clear about what this is not. The limit lives on the account, not on the key: there is no separate budget per key. If you want to cap one agent, give it its own account with its own small balance rather than your main one.

Keys you can take back

  • Revocable at any time. A revoked key answers 401 from that moment on.
  • Read-only keys. A key can be limited to reading, which is enough for an agent that only watches your servers and balance.
  • An expiry date. Set a number of days when you create the key, or leave it empty for one that never expires.
  • A rate limit of 60 requests a minute per key, so a loop gone wrong backs off instead of hammering the API. On 429, wait and retry with a growing delay.

Patterns that work

  • Ephemeral sandbox. Deploy a small server, run the untrusted code or the experiment, collect the output, delete the server. A few cents per run.
  • Worker on demand. The agent creates a server when a job arrives and deletes it when the job is done, instead of paying for an idle machine.
  • Scale out for a batch. Ten servers for an hour to crawl, build or test in parallel, then none.
  • A home for the agent itself. A small always-on server for the agent loop, as in VPS for bots, that deploys the short-lived ones.
  • Resume from a snapshot. Snapshot the server before deleting it and restore the same disk later on a new machine.

Sizes: cheap by default, heavier when needed

Start with the smallest plan, which is $10 a month and costs about 1.4 cents per hour. Most scripts, scrapers and small tools fit. When a job is steady and heavy, such as builds or a crawler that a shared core throttles, Dedicated CPU reserves the cores for you.

GPU servers are listed in the catalogue too, for inference or a rendering run. Whether a size can be ordered depends on the provider's capacity at that moment: the catalogue reports it for each plan, and an agent should only deploy plans that say they are available. Do not hard-code a plan id; fetch the live catalogue.

Keep the agent safe, and keep it inside the rules

  • Use a separate account, or at least a separate key, for each agent. If one is compromised, you revoke one key.
  • Do not put secrets on a sandbox that runs code you do not trust, and do not let an agent reuse one root password everywhere.
  • Always delete what you no longer need. Tell the agent to clean up in the same step that creates, and check the server list now and then.
  • You are responsible for what the agent does. The Terms apply to its servers as to yours: abuse, spam and attacks get a server suspended, whoever typed the command.

How to start

  • 1. Sign up with an email. No KYC, no ID, no phone and no card.
  • 2. Top up in crypto. Bitcoin, USDT, ETH or any of 300+ coins; the minimum top-up is $15.
  • 3. Create an API key in the dashboard and copy it; it is shown once.
  • 4. Give the agent the key and [/llms.txt](/llms.txt), or point it at the API docs.
  • 5. Watch the balance for the first runs and see what one job really costs.

FAQ

Can an AI agent rent a server by itself? Yes, through the REST API with an API key. A person creates the account and the key once; the agent deploys, polls and deletes servers on its own from then on.

What is llms.txt? A plain-text file at /llms.txt that describes the whole API: the steps, the endpoints, the errors and the rules an agent should follow. A model can read it in one request.

How do I stop an agent from overspending? Fund the balance only with what you are willing to lose. Servers are paid from that balance by the hour, a deploy is refused without funds, and there is no month-end invoice. The limit is per account, so give a risky agent its own account.

What happens when the balance runs out? A server the balance can no longer cover is archived: its disk is snapshotted, the server is deleted and the snapshot is kept for 7 days. Bare metal has no snapshot and is deleted with its data. Top-ups are final, see the Refund policy.

Can I give an agent a key that cannot spend money? Yes. A read-only key can list servers and read the balance but cannot deploy, stop or delete anything.

Is there a limit on how fast an agent can call the API? 60 requests a minute per key. Over that you get 429, and the agent should wait and retry with a growing delay.

Do you support GPUs for agents? GPU sizes are listed in the catalogue and open for ordering when the provider has capacity. See GPU servers for what is available now.

Can I pay without a card and without identity checks? Yes. You sign up with an email and pay with Bitcoin, USDT, ETH or any of 300+ coins. See crypto VPS hosting.