GitHub Codespaces Alternatives for Teams and Coding Agents
Eli Mernit11 min read
The right GitHub Codespaces alternative depends on what you used a codespace for. For self-hosted dev environments, the usual picks are Coder and DevPod. For a managed workstation inside your own cloud account, it is Google Cloud Workstations. For coding agents that have to keep working after you close the editor, they are a Beam VM or a Daytona sandbox. Gitpod is now Ona and sells an agent platform.
This guide compares them on where they run, what happens when they sit idle, and what the same machine costs for a working month.
What GitHub Codespaces does well, and where it stops
GitHub Codespaces gives every repository a cloud dev environment defined by files in the repo. Each codespace is a Docker container on a virtual machine that GitHub hosts, sized from 2 cores, 8 GB of RAM and 32 GB of storage up to 32 cores and 128 GB of RAM. It is the natural choice when your code already lives on GitHub and people are the ones typing.
What it does well:
- Setup lives in the repository. A dev container configuration committed to the repo gives every contributor the same tools and runtimes.
- Prebuilds. GitHub can build the environment ahead of time so a new codespace opens ready to use.
- Three ways in. You can connect from the browser, from Visual Studio Code or from the GitHub CLI.
- A free quota for personal accounts. GitHub Free includes 120 core-hours and 15 GB-months of storage a month, which is 60 hours on the 2-core machine. GitHub Pro includes 180 core-hours and 20 GB-months.
Where it stops:
- Sessions end. A codespace stops after 30 idle minutes by default, the idle timeout can be set to at most 240 minutes, and a codespace has a maximum lifetime of 12 hours even while you're using it. Stopping a codespace stops its running processes.
- Organizations get no free quota. The included hours apply to personal accounts only, so a team pays from the first hour.
- Storage bills while stopped. A stopped codespace has no compute charge and keeps accruing storage at $0.07 per GB-month until it's deleted. Inactive codespaces are deleted after 30 days by default.
- You work inside the container. GitHub's docs say you have limited access to the virtual machine underneath, and the environment is always Linux.
- It runs in GitHub's cloud. Teams that have to keep source code inside their own network need something they can host themselves.
Most of those limits don't matter for a person fixing a bug in the afternoon. They matter for a coding agent that should run a long task overnight, and for a team whose security rules keep code off third-party machines.
GitHub Codespaces alternatives compared
Coder and DevPod replace Codespaces for teams that want to host dev environments themselves. Google Cloud Workstations is the managed option inside a Google Cloud project. Beam VMs and Daytona replace it for coding agents, and Ona has moved from dev environments to background agents. The table shows what each one is, where it runs and what it does when idle.
| Option | What it is | Where it runs | When idle | Pricing |
|---|---|---|---|---|
| GitHub Codespaces | Dev containers tied to GitHub repositories | GitHub's cloud | Stops after 30 minutes by default; 12-hour maximum session | $0.18 an hour for 2 cores, plus $0.07 per GB-month of storage |
| Coder | Open-source platform for self-hosted workspaces | Your cloud or your own servers | Autostop schedules you set | Community edition free; Premium per user per year |
| DevPod | Open-source desktop app and CLI | Your laptop, your cloud account, Kubernetes or any SSH host | Set by the provider you choose | Free; you pay for the machine |
| Google Cloud Workstations | Managed workstations on Compute Engine VMs | Your Google Cloud project | Stops after an inactivity limit you set | $0.167 an hour for 2 vCPUs and 8 GiB, plus $0.20 an hour per cluster |
| Ona, formerly Gitpod | Background coding agents with managed environments | Ona's cloud or your VPC | Environments are ephemeral | Through sales |
| Daytona | Sandboxes for AI agents | Daytona's cloud | Container sandboxes stop; VM sandboxes pause with memory | $0.0504 per vCPU-hour plus $0.0162 per GiB-hour |
| Beam VM | Persistent Linux VM with SSH | Beam's cloud | Stop keeps the disk; pause keeps the disk and RAM | $0.011 per vCPU-hour plus $0.0055 per GiB-hour on the Shared tier |
A few notes on reading it. The Google Cloud Workstations figure is Google's combined compute and management price for an e2-standard-2 workstation in us-central1, and persistent disks are billed separately. The Daytona and Beam rates are per vCPU and per GiB, so the next section turns them into one machine. Beam has three CPU tiers, and the Shared tier covers VMs up to 8 vCPUs and 16 GiB.
What a cloud dev environment costs per month
For a 2-core, 8 GB machine used 160 hours a month, GitHub Codespaces costs $28.80 in compute. The table prices the same shape on each option that publishes a rate: 2 vCPUs, 8 GiB of memory, 160 hours, which is eight hours a day for twenty working days.
| Option | Inputs | Compute for 160 hours |
|---|---|---|
| Beam VM, Shared tier | 2 × $0.011 + 8 × $0.0055 = $0.066 an hour | $10.56 |
| Google Cloud Workstations | e2-standard-2 at $0.167 an hour, before the cluster fee | $26.72 |
| GitHub Codespaces | 2-core machine at $0.18 an hour | $28.80 |
| Beam VM, Performance tier | 2 × $0.05 + 8 × $0.01 = $0.18 an hour | $28.80 |
| Daytona | 2 × $0.0504 + 8 × $0.0162 = $0.2304 an hour | $36.86 |
Three things sit outside that column.
Fixed fees. Google Cloud Workstations charges $0.20 an hour for each workstation cluster whether or not anyone is working, which is $144 for a 720-hour month. One cluster serves a whole team in a region, so the fee is small per head for fifty developers and large for one.
Storage. Codespaces bills $0.07 per GB-month for the space a codespace occupies, running or stopped, so a full 32 GB disk adds up to $2.24. Daytona bills $0.000108 per GiB-hour after the first 5 GiB. Beam doesn't bill storage yet; the planned rate is $0.0001 per GB-hour, which would be about $2.30 a month for 32 GB of stored data (32 × $0.0001 × 720 hours).
Free allowances. A personal GitHub account covers the first 60 hours on the 2-core machine each month. Daytona includes $200 of free compute. Beam's Developer plan is $20 a month and includes $20 of usage credits.
Coder and DevPod have no row because the software is free and the bill is whatever your own cloud charges for the VM. Ona has no public price list.
Codespaces alternatives for coding agents
A coding agent needs three things a codespace doesn't give it: a machine that keeps running without an editor attached, a way to sleep without losing its processes, and copies of the environment for parallel attempts. Codespaces stops after its idle timeout and ends every session at 12 hours, so the options below are the ones built for unattended work.
A Beam VM is a named Linux machine with its own kernel, systemd and a root disk that persists. It runs until you stop it, and the prepared image includes the Claude Code and Codex CLIs. This creates a 2 vCPU, 8 GiB machine on the Shared tier, uploads a project and opens a shell:
# A 2 vCPU / 8 GiB dev box on the Shared CPU tier
beam vm new devbox --cpu-tier shared --cpu 2 --memory 8192
beam vm sync devbox ./my-project
beam vm ssh devbox
# Pause after 5 idle minutes, keep RAM, wake on the next connection
beam vm update devbox --ttl 300 --idle-action pause --auto-resumeThe --cpu-tier flag needs beam-client 0.2.236 or later, and sync uploads the directory's contents to /workspace. A pause saves memory and running processes along with the disk and releases compute, so a test server or a half-finished build is still there on resume. The idle timer counts connections such as SSH sessions and SDK calls, not commands running inside the VM, so leave it unset for an agent that works unattended, or hold a keep-alive lease from the SDK while the job runs.
To hand a task to an agent on the VM and walk away, run it in the background. The command prints a session ID such as prompt-2136acad6220, and you pass that ID to logs to read the output later:
beam vm prompt devbox "Run the tests and fix the failing case" --agent claude --background
# prints a session ID, for example prompt-2136acad6220
beam vm logs devbox --session prompt-2136acad6220 -fbeam vm prompt needs agent credentials, and they are yours: sign in to Claude Code or Codex inside the VM, or set your own API key there, before the first prompt. Beam doesn't supply a model subscription or a key. beam vm fork devbox devbox-experiment boots a second VM from a copy of the root disk, which gives each parallel attempt its own machine. Our guides to running Claude Code unattended and running Codex in a VM cover permissions and isolation for each agent, and the coding agents page shows the full setup.
Daytona is the other option built for this. A sandbox is a container by default, and its Linux VM sandboxes add fork, pause and memory snapshots. It bills per second at the rates in the table above. Our Daytona alternatives and Daytona pricing guides go deeper.
Ona covers the same job from the other direction: it runs the agents for you and sells the platform through its sales team.
What a Beam VM doesn't replace: it has no browser editor, it doesn't read a dev container configuration, and it has no prebuilds. It does have browser access: a new VM comes with a terminal in the browser and, by default, a Linux desktop in the browser, next to SSH and file sync. VMs are CPU-only. If your team's day is opening a pull request in a web IDE, a codespace or one of the self-hosted options below is the closer match.
Self-hosted GitHub Codespaces alternatives
Coder and DevPod are the two open-source ways to run Codespaces-style environments on machines you control. Coder is a server your platform team operates for everyone. DevPod is a client each developer installs, with no server at all.
Coder: a self-hosted workspace platform
Coder runs on your own infrastructure and provisions a workspace for each developer from a template. Templates are written in Terraform, so a workspace runs on whatever infrastructure the template defines. Developers connect from VS Code, JetBrains IDEs, Cursor or a web IDE.
The Community edition is free and open source, with unlimited workspaces. Premium is priced per user per year and adds high availability, audit logging, multi-organization access control and support. Coder also sells an agent product that runs coding agents inside the same governed workspaces.
Choose Coder when a platform team is ready to operate it and the requirement is that code never leaves your network. The cost is the team's time plus the cloud bill for the workspaces.
DevPod: Codespaces without a server
DevPod is an open-source desktop app and CLI from the maintainers of vcluster. It reads the same devcontainer.json standard that Codespaces uses and creates the environment through a provider: local Docker, Kubernetes, any machine reachable over SSH, or a VM in AWS, Google Cloud, Azure or DigitalOcean. It supports VS Code and the JetBrains IDEs, and other editors connect over SSH.
Because DevPod is client-only, there is nothing to operate and nothing to pay for beyond the machine. Each developer manages their own environments. It suits individuals and small teams that already have a dev container configuration and want to keep it.
Google Cloud Workstations: managed, inside your project
Google Cloud Workstations runs each workstation on a Compute Engine VM with a persistent disk, inside your own Google Cloud project and network. Google manages the control plane and you set an inactivity limit that stops idle workstations. Pricing is the VM, a management fee of $0.05 per vCPU-hour while the workstation runs, and the $0.20 an hour cluster fee.
It fits teams already on Google Cloud that want managed workstations under their own network controls and are large enough to spread the cluster fee.
What happened to Gitpod
Gitpod is now Ona. The gitpod.io address redirects to Ona's site, which describes the product as background agents running in the cloud with managed environments, deployed in Ona's cloud or your VPC. Its pricing page is a contact form for sales, with no self-serve price list. Ona is now part of OpenAI, which announced the acquisition after the rebrand.
If you came here looking for the old Gitpod, a hosted dev environment you could start from a repository link with your own card, the closest matches are Codespaces itself, DevPod pointed at your own cloud account, or a persistent VM you SSH into.
How to choose a GitHub Codespaces alternative
Pick by who does the typing and where the code is allowed to run.
- People, code on GitHub, no hosting rules: stay on Codespaces. The repository integration and prebuilds are the product.
- People, code must stay in your network: Coder if a platform team can run it, DevPod if each developer manages their own.
- People, already on Google Cloud: Google Cloud Workstations.
- A coding agent that runs for hours or overnight: a Beam VM, or a Daytona VM sandbox.
- An agent or a dev box you use a few hours a day: a VM that pauses with its memory and bills per second, so idle hours cost no compute.
- Agent-written code you don't trust: an isolated machine per task. Our explainer on what a microVM is covers the isolation, and AI agent hosting covers the choice between a VM, a sandbox and a container platform.
FAQ
What is the best alternative to GitHub Codespaces?
It depends on the reason for leaving. Coder is the usual pick for self-hosted team environments and DevPod for a free, client-only tool that reuses your dev container configuration. For coding agents that run unattended, a persistent VM such as a Beam VM or a Daytona sandbox fits better than any hosted IDE.
Is there a free alternative to GitHub Codespaces?
DevPod and Coder's Community edition are free and open source, and you pay only for the machines they run on. DevPod can also run environments on your own laptop with local Docker, which costs nothing.
Is there an open-source, self-hosted alternative to GitHub Codespaces?
Yes. Coder is an open-source server that provisions workspaces on your own cloud or servers from Terraform templates. DevPod is an open-source client that builds dev containers on any machine or cloud account you can reach.
How much does GitHub Codespaces cost?
Compute is $0.18 an hour for a 2-core machine and scales with cores up to $2.88 an hour for 32 cores. Storage is $0.07 per GB-month and bills while a codespace is stopped. Personal accounts get 120 core-hours and 15 GB-months free each month, and organizations get no free quota.
What happened to Gitpod?
Gitpod became Ona. The gitpod.io address now redirects to Ona's site, which sells a background-agent platform through its sales team and lists no self-serve prices. Ona is now part of OpenAI.
Can a coding agent run in GitHub Codespaces?
It can for short sessions. A codespace stops after its idle timeout, which is 30 minutes by default and 240 minutes at most, and every codespace has a 12-hour maximum lifetime. Stopping it stops the agent's processes, so long unattended jobs need a machine without those limits.
How long does a GitHub codespace stay running?
Until it has been idle for the timeout period or has reached 12 hours, whichever comes first. A stopped codespace keeps its files and is deleted after 30 days without use unless you change the retention period.
Give your coding agent its own machine
A Beam VM keeps an agent's disk, processes and memory between sessions, pauses when nobody is connected and bills per second while it runs. Create a Beam account and run the commands above to try a dev box next to your codespace. Current rates are on the Beam pricing page.
