Use it when
Your agent needs dangerous credentials.
Give a sandboxed agent scoped, temporary access to SMTP, ticketing, storage, or internal HTTP systems without handing it master keys.
self-hosted agent sandboxes
SLOPBOX
Safely run agentic workloads on your own hardware.
a bounded sandbox, launched in seconds
Keep the agent inside a box you control.
Slopbox is a self-hosted runtime for running agentic workloads on Docker or Kubernetes. Each agent gets a short-lived sandbox with scoped repositories, controlled model access, and the runtime hooks needed for real work.
A secret proxy gives agents ephemeral credentials so your long-lived keys never enter the sandbox. Inject skill files and brokered data sources, route models through LiteLLM, and keep cost tracking and observability built in.
$
slopbox
who is this for
Not another coding workspace.
Slopbox is for teams wiring agents into real operations: mailboxes, ticket queues, private data, internal APIs, and scheduled jobs.
Use it when
You already run n8n, OpenWebUI, or LibreChat.
Slopbox puts a small launch API in front of Docker or Kubernetes, so existing tools can start a contained workload on demand.
Skip it when
You just want a coding agent.
For normal development work, use devcontainers or the managed environments from OpenAI, Anthropic, AWS, GCP, or your IDE.
Different from
Raw pod sandboxing.
Kubernetes sandbox controllers isolate workloads. Slopbox adds the operator-facing API, config file, model routing, health checks, and secret proxy around that runtime.