DigitalOcean has unveiled DigitalOcean Managed Agents in public preview, introducing a managed cloud infrastructure layer designed specifically for AI agents. The new offering provides isolated microVM runtimes, governed tool access, and serverless AI inference capabilities, according to a report published by InfoQ. The platform addresses what DigitalOcean describes as the unique infrastructure demands of agentic workflows, which differ substantially from traditional cloud-based applications.
The platform combines two integrated services: a Harness Runtime and an Action Gateway. The Harness Runtime operates on lightweight microVMs, delivering persistent, isolated compute environments for agents across multiple supported frameworks including coding agents like Claude Code, Codex CLI, and OpenCode, general-purpose agents such as Hermes, and custom agents built with LangGraph. Sessions can preserve conversational history and working state across interactions, with the ability to pause, resume, or fork as needed. The runtime automatically pauses agents during idle periods when no ongoing LLM or tool calls are active, and sessions can launch in parallel across repositories and tasks to enable workflows like divide and conquer, collaboration, and map/reduce. The Action Gateway functions as the agent's interface to external tools and services through a unified MCP endpoint, currently offering access to more than 16,000 tools including Web Search, Web Fetch, Browser Automation, DigitalOcean infrastructure management APIs, and connectors for platforms like GitHub, HubSpot, and Stripe.
According to DigitalOcean, developers running agents on conventional VMs face significant challenges: "Developers are forced to invest in plumbing work to preserve the agent's context, persist artifacts and keep them accessible beyond the agent that created them, coordinate parallel work, and security-hardened access to tools." The company notes that additional obstacles include maintaining spare VM capacity to ensure agents can start quickly, as well as provisioning and configuring compute resources on demand. The new managed service aims to let developers scale the work their agents can do while DigitalOcean handles execution, persistence, tool access, and underlying infrastructure. Former Meta AI engineer Elvis Saravia, now building Dair.ai, commented on the announcement that he appreciates "the focus on performance and cost reduction, since many other agent management platforms are cost-prohibitive and make it hard to scale agents in production."
The platform's architecture is built to handle the distinctive requirements of agentic workflows. Each session runs on security-hardened compute and storage resources while allowing developers to connect to internal services without public exposure. The Action Gateway implements centralized permission management to define which tools and actions are available to each agent and requires human approval for sensitive operations. To support multiple agents accessing external systems efficiently, the gateway provides rate limiting, retries, backoff, and timeouts. Ryan Martin, platform architect and founder of Seaotter, observed that "Pause-when-idle + governed tools is the ops pattern that scales." The InfoQ report notes that DigitalOcean Managed Agents arrived around the same time as Docker's Cloud Sandboxes, with significant overlap at the sandbox and runtime level, though DigitalOcean's offering focuses more on production infrastructure with greater orchestration and tool support, while Docker Cloud Sandboxes emphasizes developer workflows and the ability to move sandboxes between local and cloud compute. The managed approach removes the infrastructure burden from development teams, letting them concentrate on agent capabilities rather than runtime plumbing. For organizations evaluating agent deployment at scale, the choice between building custom infrastructure and adopting managed platforms increasingly hinges on whether engineering time is better spent on differentiation or operational foundations.