Isolated compute environments for AI agents to run code, persist state and files, and snapshot or fork sessions.
Render Sandboxes provides isolated Linux environments for agents and other workloads to run code, edit files, and use tools. Each sandbox runs in a Firecracker microVM and can be managed through Render’s TypeScript or Python SDKs or its CLI. Users can upload and retrieve files, run commands with streamed output, and save filesystem or runtime snapshots to restore or fork environments. Sandboxes can automatically suspend when idle and resume on later activity. Network policies can restrict outbound connections; credential brokering and scoped API keys are described as early-access features.
