What can I run in a sandbox?

Run Linux tools, agent frameworks, data-processing scripts, or an application with its dependencies. Agents can install packages, edit files, execute code, and inspect the results.

Can agents keep their work between sessions?

Yes. An idle sandbox suspends, then resumes with its files intact when your agent sends its next command. You can also save a snapshot and start a new sandbox from it later.

Can multiple agents work from the same starting point?

Yes. You can start independent sandboxes from the same custom image or snapshot. Each one gets its own copy of the filesystem, so agents can run tests or work on different inputs in parallel.

Does my application need to run on Render?

No. You can create and manage sandboxes with the SDK, CLI, or API from an app hosted anywhere.

Can I run Docker Compose inside a sandbox?

Yes. Run your app, database, and other dependencies together for testing. The containers share the sandbox’s resources and lifetime.

Do snapshots preserve running processes?

Yes, runtime snapshots do. They capture the filesystem, memory, and running processes, and they restore on the same compute plan. Filesystem snapshots capture files and installed tools, and you can restore them on a different plan. Restoring a snapshot doesn’t undo changes your agent already made to outside services.

Can I restrict network access and keep API keys out of the sandbox?

Yes. You can limit outbound connections to approved domains or turn off networking entirely. Credential brokering will authenticate approved requests without giving sandbox code the raw keys. Secrets you pass in as environment variables or files stay readable inside the sandbox.

How do I debug a failed agent run?

Stream stdout and stderr, check the command’s exit code, and download generated files through the SDK or the Render CLI. Save a snapshot before terminating the sandbox if you need to investigate later.

Do I pay while my agent waits for input?

Not for CPU or memory once the sandbox suspends. By default, a sandbox suspends after 15 minutes with no commands or file transfers, and it resumes when your agent sends the next one. While it runs, you pay for the CPU it actively uses and the memory allocated to it. Snapshot storage is billed separately.

Render Sandboxes

Beta

A computer for every agent

Give every agent its own isolated compute, persistent state, and filesystem.

Create a sandbox Read the docs

TypeScriptrun-task.tsStarting sandbox

const sbx = await sandboxes.create({ timeoutSeconds: 300 });await sandboxes.upload(sbx.id, "/app/task_51.json", task);const run = "python agent.py --task 51 && pytest tests/";for await (const e of await sandboxes.exec(sbx.id, run)) {  console.log(e);}await sandboxes.snapshots.create({ sandboxId: sbx.id });await sandboxes.terminate(sbx.id);

sbx-vf1dhy95t294f3t9402yStarting

EventsDetailsSnapshots

Created00:00:00
Upload/app/task_51.json00:00:01
Executedpython agent.py --task 51 && pytest tests/00:00:01
Snapshot captured00:00:23
Completed and terminated00:00:24

Designed for you and your agents

Manage your sandbox’s lifecycle with TypeScript or Python SDKs or CLI.

Read the quickstart

const sandbox = await sandboxes.create({ timeoutSeconds: 900 });await waitForSandbox(sandbox.id);

Use cases

Each Render sandbox runs in its own Firecracker microVM, so agents can use Docker in it and resume with running processes intact.

Run your agent inside a sandbox

Give your agents tools, files, and processes that persist between sessions.

Spin up a full copy of your app

Run Docker Compose, which brings up your app and other services it needs, like a database, so your agent can test changes end to end.

Safely execute untrusted code

Execute code from users or a model, or process uploaded files, in a microVM isolated from your app’s data and secrets.

What customers say

“We’ve long been fans of Render because it enables us to move fast, and Render Sandboxes have kept that promise.”

Stuart LaytonHead of Technology, FinQore

Scale to thousands of sandboxes reliably

10,412running now

+1,204 started · −1,183 stopped in the last minute

8:0010:0012:0014:0016:0018:0020:00

Running
Suspended
Starting

= 500 sandboxes

Give your agents a head start

Start new sandboxes from custom images or saved snapshots, with the files and tools each task needs.

Stream logs and errors

Stream stdout, stderr, and exit codes so your agent can inspect a failure and try a fix.

Stay fast when demand spikes

Start sandboxes in under a second, even when traffic arrives in bursts.

Don’t pay for idle compute

Render Sandboxes suspend automatically when idle, with no compute charges until your agent sends its next command. While a sandbox runs, you’re billed only for the CPU it actually uses.

Explore pricing

Control what your agents can access

Limit what each API key can do
Create Render API keys with only the permissions a task needs, like managing sandboxes without access to other services.
Early access
Restrict outbound connections
Limit outbound requests to approved domains, or block outbound traffic entirely.
Broker sensitive credentials
Authenticate approved API requests without giving sandbox code the raw keys or tokens.
Early access

api.example.com

Create your first sandbox

Copy this prompt into your coding agent to start a sandbox, run a command, and see the output.

Follow the quickstart at https://render.com/docs/sandboxes to run my first Render sandbox. Show me the output, and terminate the sandbox when you’re done.

Agent

Read the Sandboxes quickstart
Install the Render CLI
Sign in to Render · approved in your browser
Create a sandbox · running in 0.9s
Run echo "hello from Render"hello from Render
Terminate the sandbox

Your sandbox ran the command and shut down.

Frequently asked questions

Give your agents a computer