The Cloud for AI Agents

What is The Cloud for AI Agents

Hopx provides secure, blazing-fast Linux micro-VMs designed for AI agents and untrusted code execution. It allows users to spin up isolated Linux micro-VMs in milliseconds, offering unlimited runtime and secure execution environments for various programming languages including Python, JavaScript, and Go.

How to use The Cloud for AI Agents

1. Install the HopX SDK: Use npm install @hopx-ai/sdk for JavaScript.
2. Create a sandbox: Instantiate a Sandbox object, specifying a template (e.g., 'code-interpreter').
3. Execute code: Use the runCode method to execute scripts within the sandbox.
4. Manage files: Interact with the sandbox's filesystem for uploading and downloading files.
5. Execute commands: Run shell commands and capture their output.
6. Stream output: Receive real-time output from code execution via WebSocket.
7. Manage processes: Start, monitor, and manage background processes within the sandbox.
8. Cleanup: Terminate the sandbox using the kill method when no longer needed.

Features of The Cloud for AI Agents

• Speed: Milliseconds startup time (~100ms cold start) due to prebuilt microVM snapshots.
• Security: Isolated at the VM Level using Firecracker microVMs, providing kernel isolation.
• Runtime: Unlimited execution time with full state persistence.
• SDKs: Available for Python, JavaScript/TypeScript, Go, .NET, Java, and PHP.
• Code Execution: Run Python, JavaScript, and other languages with rich output capture.
• File Operations: Upload, download, and manage files within the sandbox.
• Command Execution: Execute shell commands and capture stdout/stderr.
• Real-time Streaming: Stream output via WebSocket.
• Process Management: Start, monitor, and manage background processes.
• Templates: Predefined environments like 'code-interpreter' and options for custom templates.
• Desktop Automation: Control GUI, mouse, keyboard, and browser interactions.
• Metrics: Monitor CPU, memory, network, and disk usage in real time.
• Persistence: Long-running jobs and continuous execution without arbitrary timeouts.
• Orchestration: Compatible with frameworks like LangGraph and AutoGen for multi-agent systems.

Use Cases of The Cloud for AI Agents

• AI Agents: Running autonomous agents that write and execute code in dedicated runtimes.
• Untrusted Code Execution: Safely executing user-submitted or LLM-generated code.
• LLM Execution: Running LLM-generated code and dynamic workloads.
• Data Analysis: Launching Jupyter notebooks with preinstalled ML libraries.
• Deep Research: Enabling autonomous agents for continuous data gathering and analysis.
• Desktop Automation: Controlling cloud desktops for GUI automation and application interaction.
• Background Automations: Running workers, schedulers, and recurring jobs.
• Reinforcement Learning: Training and evaluating RL agents in isolated environments.
• Secure MCP Servers: Hosting MCP servers and tools in controlled, isolated perimeters.
• Long-Running Jobs: Executing tasks that require extended runtime without interruptions.
• Multi-Agent Mesh: Coordinating specialized agents across a network of isolated micro-VMs.

Pricing

Pricing is based on usage, with costs per second for compute, memory, and storage.

• Compute: $0.00001400/s (vCPU)
• Memory: $0.00000450/s (GiB)
• Storage: $0.00000003/s (GiB)

A free tier with $200 in credits is available for new users.

FAQ

• What is a Bunnyshell Sandbox? Secure, isolated Linux micro-VMs that launch in milliseconds, providing a runtime environment for code execution and AI agents.
• How is it different from containers or serverless functions? MicroVMs offer stronger isolation at the VM level, beyond what containers provide, and persistent runtimes unlike typical serverless functions.
• Do I need a credit card to get started? The website mentions a free $200 credit offer, implying a credit card might be required for signup or to access paid tiers after the free credits are used.
• What happens if a sandbox fails or shuts down unexpectedly? The platform aims for stability with persistent state and options for relaunching or snapshotting.