What is a cloud development environment? CDEs explained for 2026

A cloud development environment runs your dev setup on a remote container or VM. How CDEs work, Codespaces, Coder, Ona and others compared, and how to choose.

By the pyrovue team · 1 November 2026 · 10 min read

A cloud development environment (CDE) is a complete development environment that runs on remote infrastructure, a container or virtual machine in a data centre, instead of on your laptop. You open it in a browser or connect your local editor to it. The source code, the toolchain, the language servers, the builds and the running app all live on the remote machine; your laptop shows the editor and forwards ports.

People also call it a cloud dev environment, a cloud coding environment or a remote development environment. GitHub Codespaces, Coder and Google Cloud Workstations are typical examples. The environment is described in a file kept in the repository, usually devcontainer.json, so every developer gets the same one in minutes and can throw it away and create a fresh one.

The market changed a lot between 2024 and 2026. Gitpod renamed itself Ona and was bought by OpenAI, Daytona turned into sandbox infrastructure for AI agents, AWS closed Cloud9 and CodeCatalyst to new customers, and Microsoft is retiring Dev Box. This guide covers what's on offer as of November 2026, what it costs and how to pick.

How a cloud development environment works

The building blocks are much the same across products:

Compute. Each environment is a container or a VM with its own disk. It stops when idle and keeps its files while stopped. Codespaces stops after 30 minutes without activity by default (you can set anything from 5 to 240 minutes), and a running codespace has a maximum lifetime of 12 hours.
A definition in the repo. The Development Container specification describes the image, tools, ports, editor extensions and setup commands in .devcontainer/devcontainer.json. VS Code, IntelliJ IDEA, GitHub Codespaces, CodeSandbox, DevPod and Ona support it. Coder uses Terraform templates, written by a platform team, instead.
Prebuilds. Installing dependencies on every start is slow, so products build environments ahead of time. Codespaces prebuilds run as GitHub Actions workflows and use Actions minutes and storage; GitHub suggests them once creating a codespace takes more than two minutes.
Port forwarding. When your app listens on port 3000 inside the environment, the product forwards it to your browser through an authenticated URL, or to localhost on your laptop.
An editor connection. A browser editor (usually VS Code or Code OSS); VS Code desktop through its remote extensions (SSH, Dev Containers, Tunnels); or a JetBrains IDE, where the IDE backend runs remotely and a thin JetBrains Client runs locally. JetBrains Gateway connects to Codespaces, Coder, Ona, Google Cloud and Amazon CodeCatalyst, and to any Linux machine over SSH. Terminal editors work over SSH.

A minimal definition for a Node app:

{
  "name": "web",
  "image": "mcr.microsoft.com/devcontainers/typescript-node:22",
  "features": {
    "ghcr.io/devcontainers/features/github-cli:1": {}
  },
  "forwardPorts": [3000],
  "postCreateCommand": "npm ci",
  "customizations": {
    "vscode": {
      "extensions": ["dbaeumer.vscode-eslint"]
    }
  }
}

Creating a codespace from it and working in it from your terminal, with the GitHub CLI:

gh codespace create --repo OWNER/REPO --idle-timeout 60m
gh codespace ssh
gh codespace ports forward 3000:3000

What problems a CDE solves, and what it doesn't

Onboarding. A new developer, or anyone switching projects, opens an environment that already has the right runtime versions, services and dependencies, instead of working through a setup document.
"Works on my machine". Everyone starts from the same image, so differences between laptops (operating system, CPU architecture, a stray global package) stop mattering. That holds only while someone keeps the definition up to date.
Hardware. Large builds, test suites and monorepos can run on bigger machines than a laptop; Codespaces goes up to 32 cores. Developing on an ARM Mac for x86 Linux production stops being a source of surprises.
Source code security. Code stays on servers the company controls, a lost laptop holds no checkout, and offboarding means revoking one account. It isn't a data-loss guarantee: a developer can still copy files out unless policies block it.
Agents. The same infrastructure now runs coding agents. Coder says you can run Claude Code, Codex, Cursor and other agents in its workspaces, and Ona's product is built around background agents.

What a CDE doesn't fix: you pay per hour and for storage while stopped, you need a steady, low-latency connection, someone has to maintain the configuration, and reaching internal services from the environment takes network work.

Cloud development environments compared (November 2026)

Product	What it is now	Runs on	Defined by	Free	Paid
GitHub Codespaces	Dev environments for GitHub repositories	GitHub's cloud	`devcontainer.json`	120 core-hours and 15 GB-month a month on a free personal account (180 and 20 on Pro)	$0.18 an hour for 2 cores up to $2.88 for 32; storage $0.07 per GB-month
Ona (formerly Gitpod)	Environments and background agents for enterprises; part of OpenAI since August 2026	Ona's cloud or the customer's VPC	`devcontainer.json` plus automations	No; new Core subscriptions are closed	Enterprise, by quote
Coder	Open source CDE platform for platform teams, now also for agents	Infrastructure you run: Kubernetes, VMs, any cloud	Terraform templates	Community edition (AGPL-3.0); you pay for compute	Premium, per user a year (price not published)
DevPod	Open source client that creates dev containers on a provider you choose	Docker on your laptop, a VM, Kubernetes or a cloud account	`devcontainer.json`	Free; you pay for compute	None
Daytona	Sandboxes for running AI-generated code, built for agents	Daytona's cloud	SDK and API	$200 of credit	$0.0504 per vCPU-hour plus $0.0162 per GiB-hour
CodeSandbox	Browser IDE and VM sandboxes with an SDK; owned by Together AI	CodeSandbox's cloud	Its own config or dev containers	Build plan: 40 VM-hours a month	Scale from $170 per workspace a month
Replit	Build and host new apps, mostly through its Agent	Replit's cloud	Replit's config	Starter plan	Core $20 a month ($18 billed yearly)
Google Cloud Workstations	Managed environments inside a Google Cloud project	Your Google Cloud project and VPC	Container images	No separate free tier	$0.05 per vCPU-hour management fee plus VM compute and disk, plus $0.20 an hour per cluster

Codespaces, Google and Ona figures were checked on 1 November 2026; Daytona, CodeSandbox and Replit on 17 October 2026. Prices change, so check before you budget.

The main options, one by one

GitHub Codespaces is the default for code already on GitHub: open any repository you can access, in the browser or in VS Code, with prebuilds and organisation policies. Without a dev container it opens a general-purpose image and you install and start the app yourself. Billing counts core-hours, so the free 120 is 60 hours on a 2-core machine, plus storage that accrues while a codespace is stopped. The Codespaces comparison shows where it fits next to a box that runs the repo.

Ona is what Gitpod became in September 2025, when it moved from cloud workspaces to coding agents for enterprises. Gitpod Classic's pay-as-you-go plan shut down in October 2025, and Ona's billing docs say new Core subscriptions are closed. OpenAI announced on 11 June 2026 that it would buy Ona for Codex, and the deal closed on 10 August 2026. If you used Gitpod as a personal cloud IDE, plan a move; what happened to Gitpod has the timeline and Gitpod alternatives has options.

Coder is the main choice for running CDEs on infrastructure you control, from Kubernetes to on-premises VMs, which suits regulated companies. A platform team writes the Terraform templates; developers pick one and get a workspace. The Community edition is free; Premium is sold per user per year (Coder pricing).

DevPod, from Loft Labs (now vCluster Labs), is a free client with no server: it reads devcontainer.json, creates the environment on whichever provider you add, and opens VS Code, a browser VS Code or a JetBrains IDE. It also has no central management, and its 0.7 line has shipped as pre-releases, so check release activity before standardising on it.

devpod provider add docker
devpod up github.com/microsoft/vscode-remote-try-node --ide vscode

Daytona started as an open source CDE manager and became sandbox infrastructure for AI agents in 2025. Its original open source repository was archived in October 2026. Use it to give agents a computer, not to give developers a workspace.

CodeSandbox (owned by Together AI since December 2024) is a browser IDE with VM sandboxes and an SDK, suited to prototypes, shareable examples and starting sandboxes from your own code. Replit is for building and hosting new apps, increasingly through its Agent; it imports GitHub repositories through that Agent, which is billed by effort.

Google Cloud Workstations runs environments inside your Google Cloud project and network, with Code OSS in the browser or JetBrains IDEs through Gateway. The pricing has three parts: VM compute, a management fee of $0.05 per vCPU-hour while a workstation runs, and $0.20 an hour per cluster, which runs whether or not anyone is working.

AWS and Microsoft. AWS Cloud9 has been closed to new customers since 25 July 2024; existing customers can keep using it, and AWS points newcomers to its IDE toolkits and CloudShell. Amazon CodeCatalyst, which included Dev Environments, closed to new customers on 7 November 2025. Microsoft Dev Box entered its closing-down period on 14 September 2026 and retires on 18 September 2028, with Windows 365 Cloud PCs as Microsoft's recommended replacement.

What a CDE costs, and the free options

Per-hour pricing adds up in a predictable way. Two worked examples for one developer working 8 hours a day for 21 days (168 hours):

Codespaces, 4 cores: 168 × $0.36 = $60.48 a month, plus storage (a 20 GB codespace kept all month adds $1.40). The free allowance of 120 core-hours covers 30 of those hours.
Cloud Workstations, e2-standard-4 (4 vCPU, 16 GB) in us-central1: about $0.334 an hour including the management fee, so about $56 a month, plus disk. The cluster adds $0.20 × 730 hours, about $146 a month, shared by everyone on it.

With Coder and DevPod the software is free and you pay for the VMs or cluster nodes, plus the time of whoever runs the platform.

If you're looking for a cloud development environment for free, the realistic options are the Codespaces allowance on a free GitHub account (about 60 hours a month on 2 cores), CodeSandbox's Build plan (40 VM-hours a month), Replit's Starter plan, and Coder or DevPod on hardware you already pay for. pyrovue's free Hobby plan gives 20 box-hours a month, but it's a different kind of tool, explained next.

A CDE or a box that just runs the repo

A CDE is where you develop for days: a persistent workspace, your editor and extensions, and a configuration someone maintains. Plenty of jobs aren't that. You want to see whether a library works before adopting it, check a pull request running, review a candidate's take-home, or show a demo to someone who won't install anything. For those, a CDE makes you do the setup it was meant to save, because a repo without a dev container opens in a general-purpose image and leaves the install to you.

pyrovue is built for that second job. Paste a GitHub or gitlab.com link and it reads the repo's package files, README, Dockerfile, Compose file and devcontainer.json (post-create and post-start commands, forwarded ports, environment), installs and starts the app, and gives you a live preview, a terminal and an editor; most repos are running in about a minute, with no config. Free boxes sleep after 15 minutes with no viewer and are deleted a day later. It isn't trying to be a full IDE: you get a browser editor and a terminal rather than VS Code with your extensions, and there's no Docker inside the box and no GPUs. See running a GitHub repo online for how it works.

	Cloud development environment	Run-it box
Used for	Writing code for days or weeks	Trying, reviewing or demoing a repo
Setup	A dev container or template someone maintains	None for most repos
Lifetime	Persistent until you delete it	Disposable; sleeps and is deleted
Editor	Full VS Code or JetBrains, with extensions	Browser editor and terminal

How to choose a cloud development environment

Decide where code and machines must live. If a vendor's cloud is acceptable and your code is on GitHub, start with Codespaces. If they must stay in your network or accounts, look at Coder, Google Cloud Workstations (if you're on Google Cloud) or DevPod.
Decide who maintains the definition. devcontainer.json moves between Codespaces, DevPod, VS Code and JetBrains; Coder's Terraform templates need a platform team but control everything.
Check your editor. VS Code, or its open source build, works with nearly all of them. For JetBrains, confirm Gateway supports the product you pick.
Model the bill. Hours × hourly rate, plus storage while stopped, plus fixed fees such as a cluster or seats. Set idle timeouts on day one.
Check the vendor's direction. Gitpod, Daytona, Cloud9, CodeCatalyst and Dev Box all changed course or stopped taking new customers between July 2024 and October 2026. Keeping the definition in devcontainer.json makes the next move cheaper.
Ask whether you need a dev environment at all. If the goal is to try, review or demo a repository rather than develop in it, a box that runs it is faster and cheaper. The alternatives overview sets out which tool fits which job.