About

Every agent deserves a computer. Nobody should pay for it to sleep.

AI agents need a real machine: root, Docker, a network, a disk that remembers. But they work in bursts. Today you either pay for every second the machine exists, or pay a premium for every second it is active. We think agent infrastructure should cost what the work costs.

How we build

Five principles we use to decide what ships.

01

Price is a feature

Every technical decision has to lower the cost of a VM slot or raise how well we use our servers.

02

State always comes back

Losing data inside a VM is our highest-severity incident, more important than any new feature.

03

API first

The dashboard and CLI are thin layers over the public API. Nothing works only in the UI.

04

Compatibility beats originality

If copying an established resource shape shortens your migration, we copy it. An E2B-compatible adapter is planned.

05

Measure in public

We plan to publish cold-start, resume and uptime figures on a public status page, starting with the beta.

EU

EU-based team

We are based in the EU and run on dedicated servers in the EU. Data residency is the default, not an add-on.

Roadmap

Now, next, later

Status as of November 2026. "Later" items are intentions, not commitments, and dates are not promised.

Now built, validating on hardware

The core for the private beta.

  • Firecracker VMs with jailer, cgroups, netns per VM
  • Exec (sync, background, SSE, PTY), files API
  • Pause and resume with RAM snapshots, auto-pause, wake on request
  • Fork a running VM with memory, up to 50 copies per call
  • Public HTTPS URLs per port, SSH gateway
  • Templates from snapshots and OCI images
  • TypeScript and Python SDKs, CLI
  • Desktop streaming (KasmVNC) and computer-use API
  • Scoped API keys, spend limits, HMAC webhooks, usage API
  • MCP server (local and self-hosted HTTP) and llms.txt

Next before paid beta

What platforms need to move production traffic.

  • Hosted MCP endpoint
  • E2B-compatible API adapter
  • Egress allowlists
  • Public status page with resume and uptime metrics

Later after validation

Driven by what beta users ask for.

  • WebRTC desktop streaming at 60 fps
  • AI API key proxy
  • full class with nested KVM
  • GPUs
  • Open-source core and self-hosting
  • Regions outside the EU
  • SSO and audit logs
  • SOC 2 (year two)

What we won't build

Focus is how a small team ships infrastructure that holds up.

  • Our own agents. We are infrastructure, not a competitor to our customers.
  • macOS or Windows VMs in year one.
  • Anti-bot evasion, CAPTCHA solving or residential IPs.

Talk to us

Building an agent product and paying for idle VMs? We would like to hear about your workload.

Tell us about your workload hello@task.cloud