In development

Compute for AI agents

Coding agents and other AI agents can now create and run machines on their own. Before you give one access, decide the limits it works inside, and make sure the platform enforces them, not the prompt.

In development. Nothing on this page is on sale yet, and there is no account to create. The tools linked here are free and work today.

What we are building

Short-lived environments for agents are part of what ANAROQ is building, next to persistent machines. The isolation technology is not chosen yet.

Agents act only inside limits their owner sets, through the same policies as people. That is the difference we are building for; having MCP is not.

How to choose

What to check before you commit, and which tool on this site answers each question.

  1. A hard budget

    The platform stops creating resources at the limit. A budget written in the agent's instructions is a request, not a limit.

    No tool for this step: ask the provider and read their terms.

  2. A time to live on everything

    Every resource an agent creates should expire unless someone keeps it, so a forgotten run does not keep running.

    No tool for this step: ask the provider and read their terms.

  3. A cap on concurrency

    Limit how many machines can run at once, so a loop in the agent cannot multiply them.

    No tool for this step: ask the provider and read their terms.

  4. Approval for risky actions

    Deleting data, opening access or spending above a threshold should wait for a person to approve it.

    No tool for this step: ask the provider and read their terms.

  5. An audit trail and a kill switch

    Record who did what, through which interface and when, and keep one control that revokes all agent access at once.

    No tool for this step: ask the provider and read their terms.

  6. A machine with narrow access

    Give the agent a server with key-only SSH, no password or root login, and no membership in the docker group unless it needs it.

    Tool: cloud-init Generator for Ubuntu 24.04

  7. Check what the agent wrote

    Before running a Compose file an agent produced, check which ports it publishes, what privileges it asks for and which images it pulls.

    Tool: Docker Compose Analyzer

What this page does not cover

No platform on this site runs agents or gives them access to anything. The tools here produce files and reports; you decide what runs, on a server you control.