Sandbox templates allow you to customize the sandbox environment to your needs.
Novita Agent Sandbox service has built-in public templates for out-of-the-box use. You can find these public templates in the console.
To create a sandbox template, you specify the e2b.Dockerfile. We then take this Dockerfile and create a new sandbox template from it and give you back a template ID. You can then use this template ID to create a new sandbox with the SDK based on the template you created.

​
How to create custom sandbox

Steps
  1. Install E2B CLI
  2. Initialize sandbox template
  3. Customize e2b.Dockerfile
  4. Build your sandbox template
  5. Start your custom sandbox

​
1. Install E2B CLI

Please refer to the tutorial for installation and complete authentication.

​
2. Initialize sandbox template

The following command will create a basic e2b.Dockerfile in the current directory. 
export E2B_DOMAIN=sandbox.novita.ai
e2b template init

​
3. Customize e2b.Dockerfile

Now you can customize your sandbox template by editing the e2b.Dockerfile file. 
# Make sure to use this base image
FROM e2bdev/code-interpreter:latest

# Install some Python packages
RUN pip install cowsay

​
4. Build your sandbox template

Now you can build your sandbox template. We’ll use Docker and the E2B CLI. What is going to happen is that E2B CLI will call Docker to build the image and then push it to the E2B cloud. Then we convert the Docker image to a micro VM that can be then launched as a sandbox with our SDK. 
export E2B_DOMAIN=sandbox.novita.ai
e2b template build -c "/root/.jupyter/start-up.sh"
This process will take a moment. In the end, you’ll see your template ID that you’ll need to use to create a sandbox with the SDK.

​
5. Start your custom sandbox

Now you can use the template ID to create a sandbox with the SDK. 
import { sandbox } from '@e2b/code-interpreter'

// Your template ID from the previous step
const templateID = 'id-of-your-template'

// Pass the template ID to the `Sandbox.create` method
const sandbox = await Sandbox.create(templateID)

// The template installed cowsay, so we can use it
const execution = await sandbox.runCode(`
import cowsay
cowsay.say('Hello world!')
`)

console.log(execution.stdout)

​
How it works

Every time you are building a sandbox template, we create a container based on the e2b.Dockerfile file you create in the process. We extract the container’s filesystem, do provisioning and configuration (e.g. installing required dependencies), and start a sandbox. Then, these steps happen:
  1. We take the running sandbox.
  2. (Only if you specified the start command, otherwise this step is skipped) Execute the start command.
  3. Wait for readiness (by default 20 seconds if start command is specified, otherwise immediately ready). Readiness check can be configured using ready command.
  4. Snapshot the sandbox and make it ready for you to spawn it with the SDK.
We call this sandbox snapshot a sandbox template.
Snapshots are saved running sandboxes. We serialize and save the whole sandbox’s filesystem together with all the running processes in a way that can be loaded later.This allows us to load the sandbox in a few hundred milliseconds any time later with all the processes already running and the filesystem exactly as it was.