Sandbox · LINUX

Linux Container

Full Alpine or Debian environments — Docker-compatible, persistent, agent-accessible.

Linux Container sandboxes

When code-eval isn’t enough — when you need a real shell, package manager, persistent files, and the ability to keep state between calls — a Linux Container sandbox is what you want.

What this looks like

A real, isolated Linux environment. SSH-accessible (via key auth bound to the tenant), MCP-tool-accessible from the agent runtime, and persistent across calls unless you explicitly mark it ephemeral.

Sizing

  • Default: 2 GB RAM, 2 vCPU burst / 0.5 sustained, 10 GB volume
  • Upgrade: up to 16 GB RAM and 60 GB volume on per-sandbox plan
  • Egress: 200 Mbps default

What we don’t allow

  • Cryptocurrency mining (we will detect and kill it)
  • Outbound spam SMTP relay
  • Anything explicitly prohibited by Nepali or EU law

Why a Linux Container vs. a VPS

VPS is for tenant-owned long-running infrastructure billed monthly. Linux Container sandboxes are billed per CPU-second by default and are designed to be agent-controlled, not human-administered. If you find yourself running a container 24/7 for months, talk to us about converting to a VPS — usually cheaper.

Best for

  • Long-running agent worker processes
  • Build/CI-like environments where the agent compiles, runs, and reads output
  • Self-hosted MCP servers that an agent SSH/tools into
  • Multi-step workflows that don't fit in a single eval call