Software advice
Best AI Cloud IDE Software in 2026
An AI cloud IDE is a development environment that runs on a remote machine and is reached through a browser or a thin client, with AI that writes code, runs coding agents or builds whole apps inside it. It is not a local editor with a chat panel, and it is not a plain code-execution API, although one vendor here sells exactly that.
Quick answer
GitHub Codespaces ranks first for its printed free monthly allowance and native GitHub setup. Coder is the pick for self-hosted workspaces with coding agents on your own infrastructure. Replit suits people who want an agent that builds apps from prompts.
Buying a cloud IDE used to mean choosing where a developer's machine lives. The question now has a second half: which AI runs next to the code, and who controls it. Some vendors sell a hosted workspace, some sell software you host yourself, some sell agents that work in the background, and one sells sandboxes for generated code. Pricing models differ just as much, from a free allowance to hourly infrastructure meters to quote-only enterprise deals. This ranking weighs what the vendor pages print about AI features, prices, data terms, exit options and the kind of team each product suits. We read vendor pages in November 2026 and ran no tool, so claims about speed or output are the vendors' own.
The 12 best AI cloud IDE software in 2026, ranked
There is no score. Vendors can pay for more visibility on this website, for example a higher place in a ranking or more room on a page. Commercial relationships can also affect which products we feature and how prominently. How we rank.
| # | Tool | Best for | Free plan | Entry price |
|---|---|---|---|---|
| 1 | GitHub Codespaces | Teams whose code already lives on GitHub and who want a ready environment per branch | Yes | Free: 120 core hours a month on a personal account |
| 2 | Coder | Platform teams that want cloud workspaces and coding agents on their own infrastructure | Yes | Free (Community edition) |
| 3 | Ona | Enterprises that want background coding agents in cloud environments inside their own network | No | Quoted |
| 4 | Replit | Individuals and small teams who want an AI agent to build and host an app from a prompt | No | $18 per month billed annually ($20 monthly) for Core |
| 5 | Cloud Workstations | Google Cloud customers who want managed, locked-down workstations with Gemini Code Assist | No | $0.05 per vCPU per hour management fee, plus $0.20 per cluster per hour and compute |
| 6 | Lightning AI | Machine-learning practitioners who want Studios with CPU and GPU machines billed by the hour | Yes | Free: $0 with 50 free GPU hours |
| 7 | StackBlitz (Bolt) | Front-end developers who want a Node.js environment in the browser and a prompt-to-app builder | Yes | $0 per month (Bolt Free) |
| 8 | Eclipse Che | Organizations that already run Kubernetes and want an open-source workspace platform with no license fee | Yes | Free (Eclipse Public License) |
| 9 | AWS Cloud9 | AWS-centered developers who want a browser IDE tied to IAM users and EC2 | Yes | No additional charge for the IDE |
| 10 | OneCompiler | Students and instructors who run code in 150+ languages and need a browser editor for classrooms | Yes | Free forever (100 credits per day) |
| 11 | code-server | Individual developers who want to run VS Code in a browser on a server they manage | Yes | Free (open source, self-hosted) |
| 12 | Daytona | Developers building AI products that need a sandbox to run generated code through an SDK | Yes | vCPU at $0.0504 per hour, pay as you scale |
- 1
GitHub Codespaces takes first place because it is the only product here with a printed free allowance, 120 core hours and 15 GB of storage a month, and because the environment starts from the repository where the work already sits. The page does not describe AI inside the codespace itself; Copilot is a separate GitHub product. Overage rates and compliance details are not on the feature page.
Strong on
- Free personal allowance of 120 core hours or 60 hours on a 2-core machine each month
- Opens from a browser, including on an iPad, so no local setup is needed
- Organization admins can enable and pay for usage on private and internal repositories
Weak on
- The feature page names no AI capability inside a codespace; Copilot is bought and described separately
- Rates beyond the free allowance are not printed on the page opened
Skip it if: Teams hosted on GitLab or Bitbucket, or anyone who needs a self-hosted environment on their own servers. Includes 15 GB of storage. Usage above the allowance is pay-as-you-go; rates are not printed on the feature page.
- 2
Coder ranks second because it puts the workspace on infrastructure you control and adds Coder Agents, which run coding agents in ephemeral workspaces, plus an AI Governance gateway for LLM tool usage. Quotas, autostop and audit logging address cost and oversight. The price of Premium is not printed in the text read, so budgeting means asking sales, and the free tier suits only hobbyists and small teams.
Strong on
- Workspaces run on the customer's own infrastructure and accept VS Code, JetBrains, Cursor or Jupyter
- Autostop and per-user quotas limit cloud spend; SOC 2 Type II is named on the vendor site
- Coder Agents and AI Governance cover agent execution and LLM usage control
Weak on
- Audit logging, OIDC group sync and multi-organization RBAC sit in the paid Premium tier
- Someone must operate the deployment, and Premium pricing is not published in the text read
Skip it if: Solo developers who want a hosted editor with no servers to run, or teams without anyone to operate infrastructure. Premium is billed annually per user, amount not shown; AI Premium is a custom quote with a trial.
- 3
Ona sits third because it has moved from a hosted editor to cloud-run background agents for code review, migration and CVE fixes, with an option to run in the customer's VPC. That fits security-led buyers, but nothing about cost can be judged: pricing is by contact only and no free plan is named. The older Gitpod Classic service ended in October 2025.
Strong on
- Runs background agents in cloud environments, aimed at review, migration and CVE remediation work
- Can run in Ona's cloud or in the customer's own VPC
- Product pages describe governance and security controls at the kernel level
Weak on
- No public price and no free plan on the pages opened, so evaluation starts with sales
- Gitpod Classic ended 15 October 2025, so existing Gitpod users must move to the new product
Skip it if: Individual developers who want to open a quick browser editor, or buyers who need a published price before talking to anyone. Pricing is quote only; the pricing URL redirects to the sales contact page and no figure is published.
- 4
Replit takes fourth place as the clearest AI-first browser workspace with printed prices: Core at $20 a month and Pro at $100. Replit Agent builds and edits apps from natural language. The weakness for professional teams is that Pro stops at 15 collaborators and a free plan was not found in the pricing text. Training use of customer content is not stated on pages opened.
Strong on
- Replit Agent builds and edits apps from natural-language prompts inside browser workspaces
- Security page describes seccomp-bpf hardened sandboxes and an isolated Google Cloud project per customer
- Prices for Core and Pro are published, with a discount for annual billing
Weak on
- Pro allows up to 15 collaborators, so larger teams land on a custom Enterprise quote
- A named free plan was not found in the fetched pricing text; Free Mode caps sit inside Core
Skip it if: Engineering teams that need to keep code in their own cloud account or run existing devcontainer setups. Core covers personal projects with Free Mode chat usage caps; Pro is $100 monthly or $90 billed annually.
- 5
Cloud Workstations ranks fifth because the pricing is published down to the vCPU hour and the AI story is concrete: a native integration with Gemini Code Assist. The cost model has three parts, so total spend is harder to forecast than a flat seat. It also assumes you already run on Google Cloud, and no free tier appeared in the pricing text read.
Strong on
- Custom container images define the development environment, and browser or local IDEs reach the same workstation
- Gemini Code Assist integration provides AI-assisted coding inside workstations
- A cluster can connect to different VPC networks, which suits private-network access
Weak on
- A control plane fee of $0.20 per cluster per hour accrues whether or not anyone is working
- No free tier appeared in the pricing text read, and GPU machine types carry their own rates
Skip it if: Hobbyists and small teams outside Google Cloud, who would pay cluster fees for little use. The management fee runs with the workstation; the cluster fee accrues even when idle. Compute is extra.
- 6
Lightning AI ranks sixth because its Studios are built around model work: a free tier with 50 GPU hours and an hourly machine table that lists an H100 80 GB at $5.45 an hour. For general web or app development it is a narrower fit. The pricing page shows two sets of Pro and Teams prices, and the free Studio restarts every four hours.
Strong on
- Free tier gives 50 GPU hours and one Studio that restarts every 4 hours
- Teams adds a free 24/7 Studio and shared Studios; Enterprise allows bring your own cloud
- The machine table lists hourly CPU and GPU rates, including an H100 80 GB at $5.45 per hour
Weak on
- The pricing page shows two prices each for Pro and Teams, so the real figure needs confirming
- SOC 2 Type 2, HIPAA and a 99.95% SLA are listed only for Enterprise or higher tiers
Skip it if: Web and mobile app teams that want a general-purpose editor, since the product is organized around ML compute. The pricing page shows two figures each for Pro ($50 and $20) and Teams ($140 and $119 per user).
- 7
StackBlitz holds seventh place: WebContainers run Node.js in the browser with no server to rent, and Bolt turns prompts into web apps, with a free plan that is capped by tokens. Its terms grant a model-development license unless the user opts out, which a privacy-minded buyer should read first. It is aimed at web projects and ties to GitHub for saving work.
Strong on
- WebContainers run Node.js and full development environments inside the browser
- Bolt builds web apps from text prompts; Free allows 1M tokens a month
- Teams adds private npm registries and design-system prompts
Weak on
- Terms section 3.5 grants a model-development license unless the user opts out under section 3.5(c)
- Free limits tokens to 300K a day and keeps Bolt branding on sites; Pro starts at 10M monthly tokens
Skip it if: Teams working in Python, Java or other non-JavaScript stacks, or anyone who needs a long-lived server-side environment. Bolt prices: Pro $25 monthly, Teams $30 per member. Free caps tokens at 300K daily and 1M monthly.
- 8
Eclipse Che ranks eighth: an open-source cloud IDE that runs workspaces in Kubernetes pods, with a VS Code-based editor and JetBrains IDEs through Projector. The license is free and the Foundation is a named European entity in Belgium. The costs are a cluster to run and no AI capability named on the pages opened, which is a large gap in an AI-themed category.
Strong on
- Workspaces start from a Git repository or sample and run in Kubernetes pods
- A VS Code-based editor is named, and JetBrains IDEs run through JetBrains Projector
- Automated git setup runs when the workspace starts
Weak on
- No AI capability is named on the Che pages opened
- No hosted plan and no published commercial support price were found; a Kubernetes cluster is required
Skip it if: Teams that want a hosted service with AI features out of the box, or that have no Kubernetes expertise. Open-source software with no hosted price on the pages opened; the user pays for the Kubernetes cluster.
- 9
AWS Cloud9 ranks ninth: the FAQ lists no extra charge for the IDE itself and a plain feature set of preinstalled runtimes, a terminal and debugging tools, with sharing between IAM users. No AI feature inside Cloud9 was named on the pages opened, so it trails the tools that ship agents. Its FAQ describes a current service with no end-of-support date.
Strong on
- Browser IDE with preinstalled runtimes, package managers and debugging tools
- Shared environments between IAM users for real-time collaboration
- The IDE has no added charge; only the underlying AWS resources are billed
Weak on
- Each collaborator needs an IAM user, which adds setup for outside contractors
- No AI feature inside Cloud9 was named on the pages opened
Skip it if: Teams that want built-in AI coding agents, or anyone who does not already use AWS accounts. Compute and storage used by the environment are billed separately at standard AWS prices.
- 10
OneCompiler lands tenth: a free tier with 100 credits a day, code runs in 150+ languages and a Studio editor with agent features. Paid plans list credit allowances, Studio hours and named model families, but no prices were visible in the text read. Maker and headquarters were not stated on the pages opened. The free Studio lasts 45 minutes a day.
Strong on
- Runs code in 150+ languages without an install, free of charge
- Studio is a cloud editor whose agent reads the workspace and edits files
- Challenges for classrooms allow up to 60 participants on Plus and 250 on Pro
Weak on
- Free Studio time is capped at 45 minutes a day; paid plan prices were not visible in the text read
- Maker and headquarters are not stated on the pages opened, so vendor due diligence is thin
Skip it if: Companies that need a documented legal entity, compliance statements or private-network development. Paid plan prices are not published in the text read; the page lists credits, resets and Studio hours per plan.
- 11
code-server comes eleventh: it runs VS Code in a browser on a machine you control, free of charge, and pairs with Coder workspaces. It is a single-purpose tool, not a managed platform, so there is no vendor-run service, and the docs page names no AI capability of its own. Any assistant comes from the extensions you install. Only the docs landing page was read in full.
Strong on
- Runs VS Code in a browser on a remote machine under your control
- Open source and self-hosted, so there is no per-seat cost for the software
- Fits alongside Coder workspaces for browser access to the editor
Weak on
- The user installs, runs and secures the server, because no vendor-hosted service is offered
- No AI capability is named on the docs page opened; assistants depend on installed extensions
Skip it if: Teams that want managed workspaces, quotas, audit logging or built-in agents, which sit in the Coder platform. No separate price for code-server itself; the Coder pricing page offers an open-source install option.
- 12
Daytona is ranked last because it is sandbox infrastructure for running AI-generated code, not an editor you sit in. The usage-based rates are printed per vCPU, GiB and hour, with SDKs for Python, TypeScript, Ruby, Go and Java and GPU options. Buyers looking for a cloud IDE should start with the tools above; buyers building agents may start here.
Strong on
- Per-unit hourly rates are published for vCPU, memory, storage and Windows sandboxes
- SDKs are shown for Python, TypeScript, Ruby, Go and Java
- GPU sandboxes are listed, such as H100 at $3.95 and H200 at $4.54 per hour
Weak on
- It is infrastructure for running code, with no editor or AI assistant of its own named on the pages opened
- Preemptible GPU rates are lower than on-demand but can be interrupted; Enterprise pricing is sales-led
Skip it if: Anyone who wants a browser editor to type code into, since the product is an API and SDK service. Memory is $0.0162 per GiB-hour; storage $0.000108 per GiB-hour, first 5 GiB free; GPUs priced separately.
How these AI cloud IDE software were judged
Which AI cloud IDE has the strongest AI features?
The answer depends on whether you want an assistant beside you or agents working alone. Replit Agent builds and edits apps from natural language, and Bolt on StackBlitz does the same for web apps. Coder Agents and Ona run coding agents in cloud environments, and Cloud Workstations names a native Gemini Code Assist integration. GitHub Codespaces describes no AI inside the codespace on its feature page, because Copilot is a separate product. Eclipse Che, AWS Cloud9 and code-server name no AI at all on the pages opened. Test with one real ticket per tool and count how many edits the result needs.
How much does an AI cloud IDE cost?
Cost ranges from nothing to an hourly meter. GitHub Codespaces gives a personal account 120 core hours and 15 GB each month, code-server and Eclipse Che are free software, and AWS Cloud9 adds no charge for the IDE while AWS bills the compute. Replit lists Core at $20 a month and Pro at $100. Cloud Workstations meters $0.05 per vCPU hour plus $0.20 per cluster hour. Ona publishes no price and Coder does not print its Premium amount in the text read. Lightning AI shows two sets of Pro and Teams figures, so confirm the current one before you budget.
What happens to my code and prompts in a cloud IDE?
Start with who holds the machine. Self-hosted Coder, Che and code-server keep workspaces on infrastructure you pick, and Ona offers a customer VPC option. Hosted tools hold the code for you, and the terms differ. StackBlitz grants itself a model-development license unless the user opts out, which a team handling client code should read in section 3.5 of the terms. Replit names primary hosting in the United States on Google Cloud. Coder names SOC 2 Type II and Lightning AI lists SOC 2 Type 2 on higher tiers. For most others, training use and retention were not stated on the pages opened.
How hard is it to leave an AI cloud IDE?
Lock-in depends on where your environment definition lives. Tools built on repositories and container images, such as Codespaces and Cloud Workstations, keep the setup in files you can reuse elsewhere, though the billing and policy layers belong to GitHub or Google Cloud. Open-source Coder, Che and code-server can move with you, because you hold the deployment. Prompt-to-app builders such as Replit and Bolt tie projects to their own hosting and token systems, so export the code early. Ona is the reminder that a product can be replaced: Gitpod Classic ended on 15 October 2025.
Which AI cloud IDE fits my team?
Match the product to who runs it. A GitHub-based team with a few contributors can try Codespaces on the free allowance. A platform team with its own cloud and a governance brief fits Coder or Ona. Teams inside Google Cloud fit Cloud Workstations, and AWS-heavy teams can use Cloud9 for a plain browser IDE. Students and instructors can use OneCompiler for quick runs in many languages. Solo builders who want an app from a prompt fit Replit or Bolt. Machine-learning work with hourly GPU machines fits Lightning AI. If you are building agents that run generated code, Daytona is the infrastructure option.
How to choose an AI cloud IDE in three steps
- Decide who operates the environment. Choose first between a vendor-hosted workspace and one your own team runs. Codespaces, Replit and StackBlitz are hosted. Coder, Eclipse Che and code-server run on your servers, and Che needs Kubernetes. Self-hosting gives location control but adds on-call work, patching and capacity planning. If nobody on the team wants that job, rule out the self-hosted group early and compare hosted plans on quotas and per-seat cost instead. Write the answer down before you open any pricing page, because it removes half the field.
- Price one realistic month of use. Take a typical week: number of developers, hours per day each environment runs, machine size and storage. Then price it per product. Codespaces counts core hours against a free allowance, Cloud Workstations adds a per-cluster hourly fee, and Lightning AI bills hourly machines. Replit and Bolt cap usage by plan instead. Ona and Coder Premium need a quote. A tool with a low entry price can still cost more once idle clusters or token caps are counted, so compare monthly totals, not list prices.
- Check the data terms before the first repository. Read the terms for model-training language, the privacy policy for retention, and the trust page for certifications. StackBlitz states an opt-out model-development license, so a team that cannot accept that must act on the opt-out or choose another tool. Where a vendor page was silent, as with several here on retention and training, email the vendor and keep the reply. Then run a trial on a non-sensitive repository for two weeks before moving client or regulated code into any cloud environment.
Frequently asked questions
What is the best AI cloud IDE in 2026?
GitHub Codespaces ranks first here because it prints a free personal allowance of 120 core hours a month and starts from the repository teams already use. Coder is the pick for self-hosted workspaces with agents, and Replit is the pick for building apps from a prompt. The best choice depends on who operates the environment.
Is there a free AI cloud IDE?
Several have a free entry. GitHub Codespaces gives personal accounts 120 core hours a month, Lightning AI offers 50 free GPU hours, StackBlitz has a free Bolt tier, and OneCompiler gives 100 credits a day. Coder, Eclipse Che and code-server are free software, though you pay for the servers that run them. Free tiers carry caps, so check the limits.
Can I self-host an AI cloud IDE?
Yes. Coder runs workspaces on your own infrastructure, Eclipse Che runs them in Kubernetes pods, and code-server runs VS Code in a browser on a machine you control. Ona says it can run in the customer's VPC. Self-hosting gives control over location but means you maintain the deployment. Of these, only Coder names agents.
Does an AI cloud IDE train on my code?
The pages opened rarely say. StackBlitz terms grant a model-development license unless the user opts out. For Coder, Replit, Ona, Lightning AI, Daytona and OneCompiler, training use was not stated on the pages opened. For self-hosted tools it depends on the AI services you connect. Ask the vendor in writing before putting client code into any tool.
Which AI cloud IDE is best for building apps from a prompt?
Replit and StackBlitz (Bolt) are the two built for it. Replit Agent builds and edits apps from natural language, with Core at $20 a month and Pro at $100. Bolt turns text into web apps, with a free tier of 1M tokens a month and Pro at $25. Both cap usage by plan, so heavy builders hit limits.
Is Gitpod still available?
Gitpod Classic ended on 15 October 2025, according to the Ona privacy policy. The company now trades as Ona and describes background agents in cloud environments, with options to run in Ona's cloud or a customer VPC. Pricing is by contact only on the pages opened, so existing Gitpod users should contact sales.
Is AWS Cloud9 still worth using?
Cloud9 is described as a current service in its FAQ, with no end-of-support date on the pages opened, and the IDE carries no additional charge. Compute and storage are billed at AWS rates, and no AI feature inside Cloud9 was named. It suits AWS users who need a plain browser IDE, not teams after agents.
What is the difference between a cloud IDE and a sandbox service?
A cloud IDE is where a person writes code, usually in an editor opened in a browser. A sandbox service such as Daytona gives programs and AI agents an isolated machine to run code through an SDK, billed by vCPU, memory and storage per hour. Daytona is ranked last here because it is the second kind.