A sandbox development environment is an isolated workspace where you can build, break and rebuild software without touching real users, real data or your own laptop. The fastest way to get one is a hosted server you reach from a browser, sized for the job and cheap enough to throw away. Our Linux servers start at $11 per month, and our Dev & Staging hosting page shows how we frame that use.
This guide explains what the term means, how a sandbox differs from an ordinary development environment, what to look for in sandbox hosting, how to secure one, and how the setup changes for AI experiments and game servers.
A sandbox is a boundary. Encore's 2026 explainer describes it as an isolation boundary that limits what code can access in the surrounding system. Salesforce's guide puts the same idea in practical terms: a secure, isolated replica of production used to build, test and troubleshoot without affecting live systems.
That gives you three properties to look for:
In cloud computing the isolation usually comes from virtualization or containers. Salesforce notes that sandboxes commonly use virtualization or containerization to reproduce production-like infrastructure while containing code execution and limiting outside access unless it is permitted. In practice that means a virtual machine, a container, or a pool of them, each walled off from the rest.
People use the two terms as if they were rivals. They are not. Encore's explainer says a development environment is optimized for building and debugging software, while a sandbox describes how isolated the space is. A development environment can be sandboxed, and a sandbox can be used for development. The words describe different properties of the same place.
The delivery path makes the difference clearer. A university software-engineering course at Edinburgh Napier describes the development environment as where developers make frequent, not-yet-stable changes, while production is the live service end users depend on and must stay stable. Between them sit shared testing and staging stages, each with tighter control. If you want to see how that path looks in a game-development workflow, this piece on automated playtesting and what to automate shows where automated checks fit before a release reaches players.
Salesforce also separates a sandbox from a generic test environment. A sandbox is deliberately isolated from production and meant to mirror it safely, whereas general test environments may not be fully isolated or consistently updated. That is the practical line: if you cannot say what your test server can reach, it is a test server, not a sandbox.
If you want to read more on the engineering side, a curated list of programming and software engineering books recommended on podcasts is a good place to find deeper material on environments and delivery practice.
| Development environment | Sandbox | Production | |
|---|---|---|---|
| Purpose | Build and debug | Experiment safely | Serve real users |
| Stability | Changes constantly | Can be reset at will | Must stay stable |
| Data | Test or sample | Synthetic or masked | Real |
| Isolation from live systems | Varies | Required by definition | Not applicable |
Slack's engineering team describes the value of isolated development copies this way: developers can iterate and share changes for review without exposing real users or data. They also note that closer configuration parity makes a successful change more predictive of how it will behave in production.
There is a second benefit that matters to small teams and solo builders. Slack's account notes that remote environments can reduce local setup and configuration inconsistencies, especially for complex applications, whereas local development stays fast and works offline for smaller projects. If your app needs a database, a queue, a worker and a reverse proxy, a hosted sandbox saves the afternoon you would spend getting them to agree with your laptop.
Typical uses:
Here is a routine that works whether you are one developer or a team.
Write down the operating system, runtime versions, services and ports production uses. Parity is what makes a sandbox predictive. Match the OS family first (Linux or Windows), then the runtime, then the supporting services.
A virtual machine gives you a full operating system and is the closest stand-in for a traditional server. Containers are lighter and suit apps packaged with Docker. A Kubernetes cluster is the right shape when production is Kubernetes. On our platform these map to Linux servers, Windows servers and containers and clusters.
With a browser-accessible server there is nothing to install: you open a terminal or a desktop in a tab. Our servers, containers and networks are managed from one browser console, and you can run terminals and desktops side by side. That makes it simple to keep an app server, a database server and a client machine on screen together while you debug. For background on how this works, see our guides on cloud development environments and remote desktop in a browser.
Pull your repository and load synthetic or masked data. Never copy raw production records into a sandbox.
Create named accounts, use keys instead of shared passwords and close every port you do not need. Our guide to secure remote access from a browser covers the access side.
Once the sandbox is in a known-good state, keep that as your baseline. When an experiment goes wrong, you return to it rather than debug the mess.
Four questions decide most choices.
How isolated is it? Ask whether each workload has its own boundary or shares a host with strangers. We wrote up one design decision on this in per-customer Docker isolation, and compared approaches in our Docker hosting comparison.
How long does it live? Lifecycle is not built into the word sandbox. Salesforce's admin team explains that some of its sandboxes persist, while its scratch orgs are ephemeral and expire after 7 days by default, with a 30-day maximum. Decide whether you want a throwaway that disappears or a long-lived environment your team returns to, then pick hosting that fits.
Can you size it to the work? A sandbox for a quick script and one for a full app stack need different resources. Flexible per-unit sizing means you do not pay for a large machine to run a small test.
What does it cost per month? Flat monthly pricing makes budgeting easy. Our published prices:
| Need | Plan | Price |
|---|---|---|
| Single Linux sandbox | Linux Starter | $11 per month |
| Larger Linux workloads | Linux Standard / Pro | $38 / $76 per month |
| Windows sandbox | Windows Starter | $35 per month |
| Docker apps | Docker Starter / Standard | $11 / $19 per month |
| Kubernetes testing | K8s Dev Cluster | $38 per month |
| Pooled resources for several machines | Hobby Lab / Developer Studio / Team Workspace | $32 / $64 / $258 per month |
If you already have machines elsewhere, you can connect your own machines to the console at $3 per month per machine, and a public IPv4 address is $5 per month. Full details are on our pricing page.
Isolation is a safety property, not a label. Salesforce points out that code in a sandbox can still reach external systems unless that is prevented, and inadequate safeguards can leak data. Its recommendations are to limit network connections, enforce strict access controls and mask sensitive information.
Turn that into a checklist:
AI hobbyists need an environment that can run for hours, install odd packages and be wiped without regret. A sandbox gives you that, and a dedicated server means your experiments do not compete with your day-to-day machine for memory and disk.
The other half is automation. If you run your own AI agents, they need a controlled place to act. Our platform offers API and MCP endpoint access, so an agent can create, inspect and control servers through the same interface you use in the browser. Keep an agent inside its own sandbox, give it only the network access its task needs, and review what it did before anything moves toward a real system. For a small always-on agent, our Agent Micro plan is $9 per month and Agent Standard is $19 per month.
The pooled Hobby Lab plan at $32 per month suits people who want several machines for several experiments from one resource allocation.
Game servers differ from development sandboxes in one respect: they are meant to be reached by other people. So the question is not only isolation but also who can connect and how you manage the server afterward.
A sound pattern is to test mods, plugins and configuration changes on a sandbox copy first, then apply them to the live server once they work. That way a broken plugin crashes your test, not your players' session. Our game server hosting starts at $7.99 per month, and you manage the server from the same browser console as everything else. If you run a popular title, check its page, such as Garry's Mod hosting.
The differences at a glance:
| Development sandbox | AI experiment | Game server | |
|---|---|---|---|
| Who connects | You and teammates | You and your agents | Players |
| Main risk | Leaking data | Agents acting beyond scope | Broken mods, downtime |
| Lifespan | Disposable to long-lived | Often long-running | Persistent |
| Sandbox use | The whole thing | The whole thing | A test copy before changes |
Most teams treat a sandbox as something to build later, after the real servers exist. We think that order is backwards. The moment you have a server to run, you have something worth testing against, and the cheapest time to learn how it breaks is before anyone depends on it.
That is why we built one console for every machine. Linux and Windows servers, containers, networks and game servers sit in a single browser tab, with terminals and desktops side by side. There is nothing to download, so a new sandbox is minutes of decisions rather than an afternoon of setup.
We also think sizing should follow the work. A test that needs a small machine should cost like one, and a team that needs a pool of resources should be able to carve it up as it likes. Our infrastructure is owned and operated in the United States, and our API and MCP endpoint let your own AI agents work with the same servers you do.
If you already run machines elsewhere, you can connect them to the console and put sandbox and production in view together, with each kept separate.
— Servers Are Up Engineering
If you want a sandbox that opens in a browser tab, start with a Linux server from $11 per month for a single environment, or a Virtual Datacenter plan when you need several machines drawing on one pool of resources: Hobby Lab at $32 per month, Developer Studio at $64 and Team Workspace at $258.
Each plan is a flat monthly subscription, and you can add storage, networking and app hosting as you need them.
A development environment is where you build and debug software. A sandbox is an isolated space that keeps code from reaching systems it should not touch. One environment can be both, and that is usually the safest way to develop.
No. Staging is a late-stage copy of production used for final checks before release. A sandbox is for experimenting earlier, with fewer rules, and you can reset it whenever you like. Many teams use a sandbox first and staging second.
Avoid it. Use synthetic data, or masked data where sensitive fields are replaced. A sandbox should be safe to lose, and real customer records are not.
Not with a browser-accessible server. You open a terminal or desktop in your browser tab, so there is no client to install, and you can keep several servers open side by side.
Explore With AI & Cite This Page
Open a grounded prompt in your AI assistant or copy a clean Markdown, HTML, or APA reference link.
Validate protocols and subnets directly in your browser or continue reading related architecture notes.
CVE Vulnerability & CISA KEV Banner Advisor
Free DNS Zone File, CAA & DMARC Record Generator
SSH Keygen Generator & Cloud-Init YAML Builder
Reproducible Cloud Development Environments: VS Code Remote-SSH, DevPod, Agent Forwarding Risks, and KVM Sizing
Engineer remote cloud dev environments with VS Code Remote-SSH, code-server, DevPod devcontainer.json, WireGuard + ForwardAgent no security, and /srv snapshots.
Linux Server Security Hardening Checklist (2026): Ed25519 SSH, UFW, Systemd Sandboxing & Backups
Complete 10-step systems security baseline for hardening Ubuntu 24.04 LTS and Debian 12 cloud servers: Ed25519 SSH keys, default-closed inbound firewalls, Docker DOCKER-USER isolation, systemd cgroups sandboxing, and AES-256-GCM WORM backups.
24/7 Stateful AI Agent Hosting: LangGraph Checkpointing, Tool Sandboxing, SSE Streams, and VPS Sizing
Host 24/7 stateful AI agents (LangGraph, CrewAI, OpenHands, MCP, n8n, Dify): Postgres checkpointing, seccomp sandboxes, long-lived SSE, and systemd deploys.
AI Agent Infrastructure & Streamable HTTP MCP Hosting (2026): Sandboxed KVM VMs & 129 Scoped Tools
How to architect production AI agent infrastructure and Streamable HTTP Model Context Protocol (MCP) servers in 2026: KVM kernel isolation, dynamic PAT scope filtering across 129 sau_* tools, async 202 reservations, and SSE reverse-proxy tuning.