Write. Run. Test. Iterate.
Let coding agents install packages, execute generated code and run tests in their own environment.
Example task
Fix a bug, run the test suite, return the patch.
Novita Sandbox / Built for AI agents
Run code. Run CI/CD with GitHub Actions
Use browsers. Control desktops.
Give every agent an isolated runtime to turn plans into real-world tasks.
Agent task
Analyze a dataset. Build a report.
01Load data & install dependencies
02Execute Python analysis
03Save charts & report
↳ Output: report.html
¹ Average startup reported by Novita; actual latency varies by workload.
² Your account's concurrent sandbox limit follows its tier quota, shown in the pricing section below.
Where your agent gets to work
Use Sandbox when your agent needs to execute code, interact with software, or keep state across tasks.
Let coding agents install packages, execute generated code and run tests in their own environment.
Example task
Fix a bug, run the test suite, return the patch.
Run code reviews, automated tests and build validation in isolated sandboxes with GitHub Actions.
Example task
Validate a pull request and return test results.
Run Browser Use workflows to navigate pages, fill forms and extract structured results inside Sandbox.
Example task
Research product pages and export a comparison.
Run GUI workflows in isolated desktops with VNC viewing for monitoring and debugging.
Explore Computer UseProcess files, run Python analysis and save charts or reports in an isolated filesystem.
Reuse prepared environments for repeatable evaluation and reinforcement learning workloads.
Preserve filesystem and memory state with pause and resume across multi-step tasks.
The execution layer for your agent
Give each task its own runtime and filesystem. Run commands and install the dependencies your agent needs.
Pause and resume with filesystem and memory intact. Compute billing stops while paused; storage charges can apply.
Use templates for consistent starting environments and snapshots to reuse prepared sandbox state.
Access external APIs, work with files and connect browser workflows to your agent’s tools.
Pay for the resources you use
CPU and memory billed per second. No subscription required.
See full pricingStart small. Scale when you need to.
Free tier: 5 concurrent sandboxes, up to 1 hour per session.
Paid tier: 100 concurrent sandboxes. Session limits and account eligibility follow your account quotas.
Illustrative compute cost
$0.00345-minute task
1 vCPU + 512 MiB RAM
Compute estimate rounded from published rates; excludes storage and external services such as model APIs. First 60 GB persistent storage included per account. Storage is measured hourly and charged daily.
Your first agent task starts here
Explore isolated runtimes for your AI agent.
Get startedSign up as a new customer.
Choose the resources your agent needs.
Create an API key and follow the SDK quickstart to run your first task.
Open the quickstartBefore you build
An isolated execution environment where an agent can run code, use browsers, manage files and maintain state. Novita manages the runtime infrastructure so you can focus on the agent workflow.
Yes. Pausing preserves filesystem and memory state. Resuming restores the environment; killing a sandbox terminates it and it cannot be resumed. Compute billing stops while paused, while storage charges can apply.
Sandbox prices cover the allocated runtime resources. Model API calls and other external services have their own pricing. Review the relevant service prices when estimating a complete agent workflow.
The standard paid tier supports 100 concurrent sandboxes. Enterprise options offer custom resource limits and deployment regions.
Build agents that do more
Create an isolated runtime for your next agent task.