sandboxes
Most powerful & affordable cloud VMs for long-running agents
Spawn thousands of cloud computers with a browser, persistent files, for your agents, without having to manage the entire infra yourself.
curl -fsSL https://neev.ai/sandbox/install | shFull VMs in the cloud
Run Docker, systemd, databases, and cron inside. Feels like a true dev machine.
Integrated desktop & browser
To open your web apps, games, 3D scenes, IDE, and more, on a crisp 60fps virtual desktop.
Run agents inside, from outside
Prompt, observe and wrap any harness running in the sandbox, from the outside, with our CLI, API or SDKs.
Plan required, from $20/mo
- •$20 plan = $20 of sandbox time.
- •Billed per second, only while running.
- •100 to 2,000 sandboxes at once, by plan.
- •Out of time? $20 packs, auto-refilled.
4 vCPU · 8 GB RAM · 50 GB for your files
The same Node.js build & test loop, 1,000 runs, 4 vCPU / 8 GB everywhere. Time it takes:
$ sandbox newSpin up a sandbox$ sandbox ssh / scpShell in, copy files$ sandbox execRun a command, no SSH$ sandbox promptRun Claude Code or Codex$ sandbox eventsPoll agent messages & tool calls$ sandbox hostPublic HTTPS URL for a port$ sandbox desktopDesktop & browser at 60 fps$ sandbox stop / resumeSnapshot, pause billing…and many more useful features: fork, named snapshots, port forwarding, webhooks, API keys, secrets, teams, auto-stop.
Enterprise
Dedicated, private DGX Spark deployments.
Run the whole fleet on hardware that is yours alone — a private NVIDIA DGX Spark cluster, isolated from every other tenant. Your models, your data, your network. Nothing leaves the box.
Private execution
Every sandbox runs inside your dedicated DGX Spark cluster — single-tenant, air-gappable, no shared infrastructure.
Customer-owned models
Bring your own weights and serve them in-house. Your models never leave your hardware or touch a third-party API.
Controlled environment
Lock down egress, images, and secrets to match your compliance posture — SOC 2, HIPAA, or fully air-gapped.
Agent testing before deployment
Rehearse full agent runs in an isolated replica of production, so you validate behaviour before anything ships.
Enterprise sandboxes run on GPU — the same DGX Spark silicon, exposed straight to your agents for inference, fine-tuning, and GPU-bound test suites.
On the roadmap
Built for fleets of agents.
Unified MCP server
In designOne MCP endpoint that brokers every tool and data source your agents reach — managed, versioned, and audited in one place.
Secret management & credential isolation
In designPer-sandbox credential scoping, so an agent only ever sees the secrets it needs and never another tenant's.
Per-agent spend caps & access control
In designHard budget ceilings and role-based access per agent, so a runaway loop can't burn the account.
Auto-resume
ShippingStopped sandboxes wake themselves the moment a request or scheduled agent run arrives — no manual resume.
Set sail.
Spin up your first sandboxes in minutes.
Questions
Am I billed when a sandbox is stopped?+
No. Stopping snapshots the sandbox and pauses billing. Resume any time in a few seconds.
Is the $20/mo a per-sandbox minimum?+
No. It is the account total, and it converts into 555 hours of machine time to spend across any number of sandboxes.
Where do sandboxes run?+
In secure, single-tenant VMs across US and EU regions. Your data and snapshots stay in-region.
What survives stop and resume?+
Your files, installed packages, and enabled systemd services. Hand-run processes need a restart.
Can I give sandboxes to my own users?+
Yes. Create them no-env so they receive none of your account's secrets and stay confined to themselves.