Sandbox runs generated code in isolated sandboxes and sends sandbox network traffic through an egress proxy. Self-hosted deployments can run these sandboxes in Kubernetes or Docker.

Kubernetes

Deploy sandbox pods, the proxy, RBAC, and workers with the Quor Helm chart.

Docker Compose

Run sandbox containers and the proxy on a trusted single Docker host.
Use Kubernetes if more than a few people will use Sandbox, especially when they may work concurrently. Each active user needs a sandbox, so Docker Compose is best suited to small, single-host deployments.

​
Requirements

Both deployment paths require:
  • A full Quor deployment with the vector database and background workers. Sandbox does not run on Quor Lite.
  • A supported Quor release and matching application and sandbox image versions.
  • An Quor URL that the sandbox runtime can reach, configured as SANDBOX_API_SERVER_URL.
  • Capacity for a sandbox per active Sandbox user.
  • An Anthropic, OpenAI, or OpenRouter provider configured after deployment.
Sandboxes execute model-generated code. Review Sandbox Architecture before enabling Sandbox, especially the sandbox, network, credential, and sharing boundaries.

​
After the runtime is ready

1

Configure a model

In Admin Panel → Language Models, add an Anthropic, OpenAI, or OpenRouter provider and make at least one model available to your Sandbox users.
2

Configure workspace access

Open Admin Panel → Sandbox → Access. Keep the workspace default off for a limited rollout, or enable Sandbox for everyone.
3

Run a complete test

As an enabled user, start a Sandbox session, send a prompt that creates a file, and open or download the result.
4

Add Apps and Skills

Configure external services under Admin Panel → Sandbox → Apps and reusable workflows under Sandbox → Skills.

Managing Sandbox

Configure access, models, Apps, and workspace instructions.

Sandbox Architecture

Understand sandbox isolation, egress, credentials, and approvals.