Private beta · hosted in Zurich

Machines your agents can rent themselves.

ARM machines and L4 GPUs, hosted in Switzerland. Your agents start, use and pay for them over the API or MCP, from credits you add up front and within the caps you set.

Read llms.txt Works with any agent that speaks MCP or HTTP.

Two customers

Built for the agent that runs the work, and the person who pays for it.

for your agent

Everything is something it can read and call.

MCP server
Tools like create_machine, exec and get_usage, ready in any MCP client.
Typed API
REST with an OpenAPI spec, predictable errors and idempotent creates, so a retry never starts a second machine.
Machine-readable docs
An llms.txt written for models, so an agent can learn the API without a human in the loop.
No signup or console
Your agent never touches a signup or console. You create its key once.

for you

You decide how far it can go.

Per-agent spend caps
Each agent gets its own key, budget and machine limit, checked before anything launches.
A hard stop at the cap
When an agent reaches its cap or your credits run out, its machines stop. No overage, no surprise invoice.
One isolated machine per job
Every lease is its own VM, created for one job and destroyed after. Nothing is shared between agents.
A record of every lease
Which agent ran what, for how long, and what it cost, down to the second.

Pricing

Pay by the second.

Prepaid credits that never expire. Billed per second, 60-second minimum. Every machine stops at its time limit, at its cap, or when credits run out.

ARM sandbox

arm64

from $0.20/h

≈ $0.0000333 per second

Beta pricing, subject to change

2 vCPU, 4 GiB memory (ARM64 Linux)
Dedicated VM per lease
Root disk and outbound network
exec over the API and MCP
SKU arm-small

L4 GPU

add-on

from $2.00/h

≈ $0.00045 per second

Beta pricing, subject to change

1× NVIDIA L4, 24 GB
Same API, same caps, same TTL
Root disk and outbound network
Limited beta capacity
SKU gpu-l4

Estimate a job

Machine type × minutes, 60-second minimum.

Beta time limits max out at 2 hours per machine.

Prices in USD, excluding VAT where applicable, charged against prepaid credits. Top-ups start at $10. Credits never expire and aren't refundable during the beta. Beta pricing, subject to change.

Quickstart

Two minutes to a first machine.

Give your agent a key with a cap, add the MCP server, and let it work. Or call the API yourself.

Add MCP · mcp.json

{
  "mcpServers": {
    "capvm": {
      "url": "https://api.capvm.com/mcp",
      "headers": {
        "Authorization": "Bearer ak_test_YOUR_AGENT_KEY"
      }
    }
  }
}

curl

curl https://api.capvm.com/v1/machines \
  -H "Authorization: Bearer $AGENT_KEY" \
  -H "Idempotency-Key: job-42" \
  -H "Content-Type: application/json" \
  -d '{
    "sku": "arm-small",
    "ttl_seconds": 900
  }'

Endpoints

POST /v1/machines
GET /v1/machines/{id}
POST /v1/machines/{id}/exec
DELETE /v1/machines/{id}
GET /v1/usage

SKUs: arm-small and gpu-l4. Use ak_test_ keys while you build, ak_live_ keys in production.

Every machine needs ttl_seconds. During the beta the maximum is 2 hours (7200).

Docs for agents: /llms.txt and /openapi.json. Endpoints shown are placeholders until the beta opens.

Trust

Hand over the keys without handing over the wallet.

Hosted in Switzerland

Machines and the control plane are hosted in Zurich, Switzerland.

A dedicated VM per lease

Every machine is its own virtual machine, created for one lease and destroyed after.

Hard stop at zero

When credits or a cap run out, machines stop. No overage, no surprise invoice.

No forgotten instances

Every machine has a required time limit and is cleaned up when it ends, even if the agent forgets.

Let your agents get to work.

We're inviting the first testers now. Leave your email and we'll send you an API key when your spot opens.