OpenComputer

What is OpenComputer

OpenComputer is a persistent VM service for agentic workloads. It gives AI agents a real computer with a full filesystem, full OS access, and state that survives across sessions. The core idea is simple: instead of launching a throwaway sandbox for every task, OpenComputer keeps a VM around until you stop or delete it. That makes it better suited to long-running LLM agents that install tools, write files, revisit the same working directory, or continue a session later.

The platform is built around durable infrastructure rather than ephemeral compute. According to the site, each VM can hibernate when idle and wake in seconds, so you can pause work without losing the environment. OpenComputer also says it is designed to work with every agent, with a particular fit for harnesses such as Claude Agent SDK.

Key features

OpenComputer focuses on persistent VMs and agent-friendly compute controls:

• Persistent VMs: machines stay available until you explicitly stop or delete them.
• Hibernate and wake: idle VMs can hibernate and resume in seconds.
• Elastic compute: resize memory and CPU while the VM is running.
• Checkpoints: take instant snapshots, then restore or fork from a prior state.
• Full computer access: agents get a full OS and filesystem, not a constrained sandbox.
• Usage-based billing: the site says you pay only while running, with disk beyond the included amount billed separately.
• Scale for platforms: OpenComputer says it can scale to thousands in the cloud for B2B agent products.

A few details matter if you are comparing it to a standard sandbox. First, the app is not just about isolation; it is about isolation plus persistence. Second, you can resize CPU and memory without restarting the machine. Third, a bad VM state does not have to be fatal because checkpoints let you roll back quickly.

How to use it

Using OpenComputer is presented as a straightforward VM flow rather than a complex infrastructure project.

1. Log in and create a VM. The site invites you to try the product by creating your first VM.
2. Connect your agent or harness. OpenComputer says it works with every agent and highlights Claude Agent SDK as a supported style of workload.
3. Install what the agent needs. Because the VM has a full filesystem and OS access, the agent can keep tools and files between sessions.
4. Resize resources during the run. If the workload needs more CPU or memory, you can adjust it while the VM is live.
5. Hibernate when idle. When the agent is waiting, put the VM to sleep instead of destroying the environment.
6. Wake or restore later. Resume from the same state, or use checkpoints to fork or roll back to a known point.

For teams building an agent product, the value is in reducing repeated setup work. If an agent installs dependencies once, OpenComputer is meant to keep that state around instead of forcing a fresh start every time.

Who it is for

OpenComputer is aimed at developers building AI systems that need longer-lived execution environments. It is especially relevant for:

• B2B agent platforms that want users to have a durable workspace.
• Teams building AI coding or automation products similar in spirit to tools like Lovable, Devin, or Bolt.
• Agent engineers who need an environment that can remember files, packages, and session state.
• Builders using Claude Agent SDK or similar harnesses that need a machine the agent can return to.

It is less about simple one-off jobs and more about workloads that continue over time. If your agent needs to keep a repo, preserve installed dependencies, or resume after a pause, OpenComputer fits that pattern better than a disposable sandbox.

What to know before you use it

OpenComputer is designed for persistent agent compute, but there are practical limits to keep in mind.

• The site states that each VM includes 20 GB of disk; storage above that is billed separately.
• Pricing is usage-based and depends on whether the VM is on the standard or pre-booked path.
• The published pricing shows pre-booked compute as cheaper than instant, with a 30-minute notice requirement and no commitment.
• Disk over the included allowance is billed per GB-second for the lifetime of the VM, including while hibernated.
• OpenComputer is not described as a general consumer cloud desktop, and the product messaging is centered on AI agents and harnesses.
• It is also not the same thing as a simple sandbox service: if you only need a short-lived isolated script runner, the persistent VM model may be more than you need.

Because the product is built for agent infrastructure, it is best used when state continuity matters. If you need an authoritative machine image, infrastructure policy detail, or live billing estimate for a production deployment, check the docs and pricing pages directly before committing. OpenComputer is strongest when you want a persistent VM for an AI agent, but official docs remain the right source for exact limits, current pricing, and account-specific constraints.