Agent Sandboxes

Agent Sandboxes enable you to run autonomous agents in isolated, managed environments on NKS (Nirvana Kubernetes Service) clusters managed by Nirvana. Each sandbox can be paused, resumed, and configured with persistent storage — perfect for long-running agents that need to checkpoint state and continue exactly where they left off.

Features

Instant Lifecycle Management

Create, pause, resume, and delete sandboxes in seconds. Pause instantly frees compute resources while retaining all files, and resume restarts from the exact checkpoint.

Persistent Storage

Optional PVC-backed /workspace directory survives pause/resume cycles, enabling true resumable workflows for long-running agents that need to maintain state across sessions.

Gateway Access

Direct HTTP/TCP access to sandbox services via public endpoints. Each sandbox can expose multiple ports through the gateway for real-time communication with external systems.

Multi-Language SDKs

Interact with Agent Sandboxes using official SDKs for TypeScript, Python, Go, C#, Java, and Kotlin, or use the REST API directly.

High-Performance Hardware

Built on AMD EPYC 9354P CPUs (32 cores/64 threads, 3.25–3.8 GHz) with DDR5 RAM for maximum performance and reliability.

Enterprise Reliability

Hard anti-affinity ensures the cluster continues operating normally even if individual nodes fail. Your workloads are protected by Nirvana’s managed infrastructure.

Performance

Ephemeral Sandboxes:

Create: ~2.5 seconds
Pause: Instant
Resume: Instant
Delete: Instant

Persistent Sandboxes:

Create: ~6-7 seconds
Pause: Instant
Resume: Instant
Delete: Instant

Architecture

Agent Sandboxes is OpenSandbox running on Nirvana’s NKS with a custom, performance-tuned configuration.

The service is managed by the Nirvana team with:

Custom NKS Configuration — Your sandbox cluster runs on dedicated NKS infrastructure tuned specifically for your workloads
Integrated Storage Solution — ABS (Accelerated Block Storage) for persistent /workspace storage, seamlessly integrated through our extensions API
Per-Customer Performance Tuning — All cluster parameters (CPU, memory, networking, storage latency) are optimized for your specific requirements

Access Points:

API Server — Lifecycle operations (create, pause, resume, delete)
Gateway — Data plane access (exec, file transfer, service endpoints)

REST API

Manage sandboxes through a simple REST API with authentication via API key. All operations are available immediately:

Create — Spin up a new sandbox with custom image, resources, and storage
Pause — Instantly free compute resources while retaining persistent storage
Resume — Restart from checkpoint in seconds
Delete — Clean up when done
Access — Reach services inside your sandbox via gateway or API proxy
Inspect — Check logs, events, and diagnostics

Authentication is simple — include your OPEN-SANDBOX-API-KEY header with every request. See the API Reference for complete endpoint documentation.

Use Cases

Long-running agents — Agents that process tasks over hours or days with periodic checkpointing
Resumable workflows — Multi-step processes that can be paused and resumed without losing progress
Isolated execution — Running untrusted or experimental code in secure, isolated environments
Development & testing — Spinning up ephemeral sandboxes for quick testing and iteration
Batch processing — Running parallel jobs with persistent state management

Getting Started

Ready to create your first sandbox? Start with Creating a Sandbox.

For API documentation and endpoints, see the API Reference.

For persistence configuration and our integrated storage solution, see Persistence & Storage.

Early Product

Agent Sandboxes is an early release. We welcome your feedback and bug reports to help us improve the product.