The Cloud for AI Agents

Product Introduction

1. Hopx is a cloud platform designed to deploy and manage isolated Linux micro-VMs optimized for executing AI agents, untrusted code, and scalable computational workloads. It leverages Firecracker microVM technology to spin up fully isolated environments in approximately 100 milliseconds, combining the security of hardware-level virtualization with near-instant startup times.
2. The core value of Hopx lies in providing a secure, high-performance infrastructure for AI-driven workflows, enabling developers to run untrusted code, autonomous agents, and long-running jobs without cold-start delays or arbitrary runtime restrictions. It bridges the gap between container agility and VM-level security while maintaining persistent state and multi-language SDK support.

Main Features

1. Millisecond-Scale Cold Starts: Sandboxes launch in ~100 ms using prebuilt snapshots, bypassing traditional VM or container initialization delays. This is achieved through Firecracker’s lightweight virtualization and optimized kernel footprints, making it suitable for burst-scale AI agent deployments.
2. Hardware-Enforced Isolation: Each sandbox runs in a dedicated Firecracker microVM with kernel-level separation, providing stronger security than containers or serverless functions. This prevents host system exposure while allowing full Linux capabilities, including file operations, shell commands, and background processes.
3. Unlimited Runtime with State Persistence: Sandboxes operate without execution time limits, supporting continuous AI agent operations, Jupyter notebooks, or long-running jobs. Persistent storage and IPython kernel integration ensure state retention across sessions, with real-time metrics for CPU, memory, and network monitoring.

Problems Solved

1. Secure Execution of Untrusted Code: Hopx addresses the risk of executing LLM-generated or user-submitted code by isolating every workload in hardware-enforced microVMs. This prevents host system compromise while allowing full Linux functionality, unlike restricted serverless environments.
2. Scalable AI Agent Infrastructure: Developers building autonomous AI agents, reinforcement learning models, or multi-agent systems benefit from rapid scaling, per-agent isolation, and SDK-driven orchestration. The platform eliminates cold-start bottlenecks and runtime limits that disrupt agent continuity.
3. Persistent, Cost-Efficient Workloads: Data scientists and engineers can run Jupyter notebooks, background jobs, or GUI automation tools without timeouts or frequent reinitialization. Hopx’s per-second billing and microVM efficiency reduce costs compared to always-on VM clusters.

Unique Advantages

1. Superior Isolation vs. Containers: Unlike Docker or Kubernetes, Hopx uses Firecracker microVMs with dedicated kernels, eliminating shared-kernel vulnerabilities. This enables safe execution of malicious or buggy code without host system exposure.
2. Multi-Language SDKs and Desktop Automation: Prebuilt SDKs for Python, JS/TS, Go, and .NET simplify integration, while unique features like GUI automation (VNC/noVNC streaming, mouse/keyboard control) support advanced AI use cases.
3. Elastic Scaling with Zero Cold Starts: The combination of snapshot-based boot and Firecracker’s efficiency allows scaling to thousands of concurrent sandboxes with subsecond latency, outperforming traditional VM clouds and serverless platforms.

Frequently Asked Questions (FAQ)

1. What is a Bunnyshell Sandbox? A Bunnyshell Sandbox is a Firecracker microVM instance preconfigured with Linux, SDK access, and persistent storage, optimized for running AI agents or untrusted code. It provides full filesystem control, network isolation, and real-time metrics, accessible via APIs in multiple programming languages.
2. How is it different from containers or serverless functions? Unlike containers, microVMs offer kernel-level isolation, preventing host system breaches. Compared to serverless functions, Hopx supports unlimited runtime, background processes, and stateful workloads without cold starts, using pre-warmed snapshots.
3. Do I need a credit card to get started? No credit card is required for the free tier, which includes $200 in credits for testing sandboxes. Users can deploy agents, execute code, and explore SDK features without upfront payment or time limits.