DigitalOcean Launches Managed Infrastructure for AI Agents

DigitalOcean has introduced DigitalOcean Managed Agents in public preview to address the infrastructure challenges of running agentic workflows. Unlike traditional cloud applications, AI agents running on standard virtual machines require developers to build custom runtime environments, manage state persistence, and secure tool access. This new offering eliminates that overhead by combining two integrated services: the Harness Runtime and the Action Gateway.

The Harness Runtime uses lightweight microVMs to deliver isolated, secure compute environments. It supports coding agents such as Claude Code, Codex CLI, and OpenCode, alongside general-purpose agents like Hermes and custom systems built with LangGraph. Crucially, the runtime can persist conversation history and working states across sessions, allowing developers to pause, resume, or fork agent tasks. It also features an automatic pause-when-idle capability to reduce costs when no large language model or tool calls are active.

To connect agents to external resources, the Action Gateway provides a unified Model Context Protocol endpoint with access to more than 16,000 tools. These include Web Search, Web Fetch, Browser Automation, and DigitalOcean infrastructure APIs, as well as integrations for platforms like GitHub, HubSpot, and Stripe. The gateway secures these connections through centralized permission management, rate limiting, and human-in-the-loop approval requirements for sensitive operations.

For practitioners, this release shifts the burden of scaling agents from manual infrastructure plumbing to managed orchestration. The platform competes with Docker Cloud Sandboxes, which also offer microVM isolation. However, while Docker focuses primarily on local-to-cloud developer workflows, DigitalOcean targets production-grade infrastructure with robust tool integration. This setup allows developers to run parallel agent sessions for complex collaborative tasks without managing the underlying servers.