Cube SandboxEmpowering your AI Agents.
Instant, Concurrent, Secure & Lightweight Sandbox Service for AI Agents
Skip to content
Next-Gen Virtualized Sandbox Infrastructure for AI Agents
From hardware-isolated MicroVM kernels to sub-100ms Copy-on-Write memory forks, engineered for deterministic, multi-tenant agent execution.
< 60ms
Instant Cold Start
Snapshot clones bypass kernel boot overhead
< 5MB
Memory Overhead
CoW page sharing, thousands per node
1:1 MicroVM
Hardware Isolation
Dedicated guest OS kernel per sandbox
Drop-in
E2B SDK Compatible
Compatible with E2B SDK, switch seamlessly via env var
01SYS_CORE // SNAPSHOT_ENGINE
LATENCY: SUB_60ms // COLD_START
Sub-60ms Cold Start & Memory Snapshot Forking
Pre-warmed memory pooling combined with Copy-on-Write micro-clones eliminates standard kernel boot overhead. Sub-60ms instance instantiation powers instant, deterministic agent workflows.
Explore Snapshot & Clone
02SYS_ISOLATION // HARDWARE_VIRT
BOUNDARY: RING_0_DEDICATED
Hardware-level MicroVM Kernel Isolation
Every sandbox executes within its own dedicated Linux kernel inside a lightweight MicroVM boundary, physically preventing container breakouts and malicious host compromise.
Explore Architecture Boundary
03SYS_COMPAT // E2B_SDK
API: E2B_COMPATIBLE // DROP_IN
E2B SDK Drop-in Compatible
Compatible with E2B SDK interface. Switch from E2B Cloud seamlessly by changing one environment variable — zero client code changes.
Explore SDK Tutorial
04SYS_NET // EBPF_L7_GATEWAY
FILTER: TC_EGRESS // SECRET_SHIELD
eBPF Kernel Egress & L7 Credential Proxy
Kernel-level eBPF TC hooks enforce strict inter-sandbox isolation. Built-in L7 security reverse proxy manages route authorization and automatic credential injection without code exposure.
Explore Network Security
05SYS_DENSITY // MEMORY_COW
DENSITY: 1000+
MB-level Footprint & Extreme Host Density
Through shared read-only kernel structures and memory pages, per-sandbox overhead drops to single-digit MBs, supporting thousands of resident instances with auto-sleep/resume.
Explore Performance Benchmarks
06SYS_STATE // DELTA_TREE
CHECKPOINT: DELTA_TREE
State Time Machine & Tree Forking
Millisecond checkpoints capture point-in-time runtime deltas. Roll back seamlessly or fork divergent exploration paths for complex agent decision-tree validation.
Explore State Branching
07SYS_STORAGE // VOLUME_BUS
STORAGE: DECOUPLED_MOUNT
Decoupled Volume Storage Framework
Storage lifecycles operate independently from sandbox instances. Pluggable drivers support S3, host mounts, and shared block volumes with seamless hotplug.
Explore Volume Plugins
08SYS_INFRA // MULTI_NODE
TOPOLOGY: MULTI_NODE
Multi-Node Cluster Deployment
Seamlessly scale single-node sandboxes to a multi-node cluster. Centralized CubeMaster orchestrates compute nodes with high availability and minimal operational overhead.
Explore Multi-Node Deployment
09SYS_ARCH // AARCH64_NATIVE
ISA: ARM64_FULL_STACK
Native ARM64 Architecture Pipeline
Deep optimization for AArch64 instruction sets spanning the custom hypervisor, minimal kernel, and OCI image execution, delivering peak density and efficiency.
Explore ARM Native Support
Get Started
- Quick Start — from zero to a running sandbox in minutes
- Self-Build Deployment — build from source and deploy on a single machine
- Multi-Node Cluster — scale to multiple nodes
- Architecture Overview — understand the system design and core components
Examples
For SDK examples and end-to-end scenarios, see:
- Example Projects — code execution, browser automation, OpenClaw integration, RL training, and more
- Repository examples — full collection on GitHub