E2B Embed puts AI agent sandboxes inside a customer's own environment

What E2B Embed is

E2B Embed brings the E2B Runtime and its dashboard to one machine that the vendor controls. It sits next to the existing E2B Cloud and E2B Bring Your Own Cloud offerings. E2B says the reason for building it is that agent companies need to serve customers whose data has to remain inside their own environment, which is often a requirement for government organizations and companies in regulated industries.

The idea is that a company can include E2B in the software it delivers to such customers. The sandbox stack runs on one machine in the customer's environment, and the vendor operates it.

What is included and what is not

E2B Embed runs the open-source E2B Runtime that powers E2B sandboxes. The databases run on the machine, and the templates a team builds and its sandbox logs are stored on the machine's disk. According to E2B, the package includes:

• Firecracker sandboxes: every sandbox is a Firecracker microVM, using the same VM isolation as E2B Cloud.
• The same SDKs: the Python or JavaScript SDK works with the install's team API key and SDK URLs.
• Custom templates: teams can build their own with Template.build and start sandboxes from them. Builds run inside a Firecracker VM on the machine.
• A dashboard: users can browse sandboxes and templates, with a terminal and a file browser for each sandbox. It comes with every installation option.
• OpenTelemetry export: one setting sends metrics, traces and logs from the E2B services to a collector of the user's choice.

E2B Embed does not include secrets, volumes, workload identity or bring-your-own proxy. It is also a single-machine product. Teams that need to scale across machines, or want E2B to operate the deployment in their cloud account, are pointed to E2B Bring Your Own Cloud.

Installation options

E2B Embed is available in the E2B Runtime repository. There are four setup options, and all run the same stack on one machine:

• Docker Compose: needs a Linux host with KVM, installed with docker compose up -d --wait.
• Terraform on GCP: needs a GCP project and credentials, using terraform init && terraform apply.
• Terraform on AWS: needs an AWS account and credentials, using the same Terraform commands.
• Kubernetes: needs a cluster with a prepared KVM node, installed with kubectl apply -k.

For Docker Compose, the host must be a dedicated Linux machine with KVM and Docker. The install is two files and one command. On the first run, the setup checks and prepares the host, downloads the Firecracker artifacts, pulls the pinned images, migrates the databases and builds the base template. The command succeeds once that template is ready. Running docker compose logs ready shows the team API key, SDK URLs and dashboard URL.

Embed serves plain HTTP inside the network. E2B directs users to its networking guide to configure access to the API and dashboard and to keep internal service ports private.

Licensing and what it means

E2B Embed is open source under Apache-2.0, like the E2B Runtime. According to E2B, users do not need an E2B account or a license key, and contributions are welcome.

E2B frames the release around safe deployment. An agent that can run code and work with files needs clear limits on what it can access, and the company says businesses cannot rely on the assumption that an agent will always behave correctly. E2B gives agents isolated VMs to work in, though it notes that isolation is one part of safe deployment, alongside network and access controls. For agent vendors with customers in regulated sectors, Embed offers a route to ship those sandboxes into the environment the customer requires, within the limits of a single machine.