Available now

Give AI agents computers they can actually use

Isolated sandboxes for real execution

Run agent workloads with real terminals, files, services, and browsers. Use Borealis-managed capacity or qualifying private host pools, then stop, snapshot, and resume without losing work.

  • Start, stop, resume, and inspect isolated sandboxes
  • Persistent snapshots and exact workspace transfer
  • Managed capacity with plan-controlled private host pools
  • Metered vCPU and memory only while sandboxes run
Start FreeSign In

Sandbox

agent-workspace-01

Running

Compute

4 vCPU · 8 GiB

Checkpoint

2 min ago

terminal
$ pnpm test ✓ 48 tests passed $ _

Infrastructure built for agent workloads

Everything an agent needs to do real work, with the isolation and controls your organisation needs.

Isolated Execution

Give each workload a dedicated sandbox with explicit CPU, memory, networking, and lifecycle boundaries.

Persistent Snapshots

Checkpoint a sandbox, stop it when idle, and resume later without rebuilding the working environment.

Managed or Private Capacity

Run on Borealis-managed hosts or enrol private host pools when your plan and security requirements call for them.

Terminals, Logs, and Ports

Execute commands, stream logs, expose services, and inspect running sandboxes through one consistent control plane.

Usage-Based Billing

Track allocated vCPU and memory while sandboxes run. Stopped and suspended sandboxes do not incur compute charges.

API, SDK, and MCP

Create and control sandboxes from the web app or automate them with the versioned API, .NET client, and MCP tools.

From image to working environment

Create, run, preserve, and automate a sandbox in five clear steps.

Borealis

Define the Sandbox

image: node:22
cpu: 4
memory: 8Gi
ports: [3000]