Agent Compute Engine
Agent Compute Engine is a managed Google Cloud compute service that provides execution environments for AI agents. Each agent receives a hardware-isolated microVM with its own guest kernel and root-level control. The service supports rapid startup, pausing, checkpointing, snapshots, and resume workflows. Persistent workspaces retain agent files and context across sessions, while copy-on-write memory sharing supports parallel environments from common images. Agents can run code, shell commands, and browser actions in isolated sandboxes, with governed access to enterprise data through Google Cloud networking and identity controls. Developers can configure environment images, identity, network access, and lifecycle while Google manages the underlying compute fleet. The platform is designed for interactive assistants, coding agents, browser-based tasks, and large-scale evaluation and reinforcement-learning workloads.
Learn more about Agent Compute Engine
Daytona
Daytona is a cloud-native development runtime that enables developers and AI agents to instantly create, run, and manage isolated sandboxes for any codebase. Each sandbox runs inside a secure microVM with full Linux compatibility, networking, and persistent storage.
Daytona provides SDKs in Python and TypeScript, allowing applications to programmatically execute code, run processes, upload files, or spin up environments dynamically.
Teams use Daytona to replace complex local setups with reproducible cloud sandboxes that can be started in seconds and accessed through preview URLs, SSH, or APIs. It’s built for automation, observability, and scalability, powering everything from personal development environments to enterprise-grade agent runtimes.
Learn more about Daytona
CreateOS Sandbox
CreateOS Sandbox is a hosted compute service for developers building AI agents, code-execution platforms, and developer tools. It creates isolated Firecracker micro-virtual machines, each with its own Linux kernel, where teams can run untrusted or model-generated code. Developers can provision environments through a TypeScript SDK, command-line interface, MCP tools, or REST API, execute commands, and connect sandboxes through private networks. Sandboxes can be paused and resumed with guest state preserved, configured from reusable templates, and connected to customer-owned Amazon S3, MinIO, or Cloudflare R2 storage. Usage is metered per second for compute, memory, and disk; paused environments continue to incur memory and storage charges. The service offers EU and US regions, plus an enterprise option to run on customer infrastructure.
Learn more about CreateOS Sandbox
Control Plane
Control Plane is the unbreakable platform for running production workloads. It runs serverless, standard, stateful, and cron workloads natively on AWS, GCP, and Azure, VM workloads on Managed Kubernetes locations, and workloads on your own infrastructure or on premises, active-active across regions and clouds. The 99.999% availability SLA applies to workloads running at least two replicas in at least two locations.
Control Plane handles Day 2 operations for you, including autoscaling, Capacity AI right-sizing, built-in observability, and security, with no Kubernetes cluster to operate. Universal Cloud Identity gives workloads access to 600+ services across AWS, GCP, and Azure with no long-lived cloud keys in your code.
Customers typically cut compute costs 30 to 50 percent compared with running directly on AWS, GCP, or Azure. Control Plane is PCI DSS Level 1, SOC 2 Type II, HIPAA, and GDPR compliant.
Learn more about Control Plane