Summary
DigitalOcean Managed Agents provides persistent, isolated environments for deploying AI agents with governed access to tools.
About
A managed infrastructure platform for deploying persistent AI agents in isolated environments, with centralized, governed access to external tools. It may reduce the operational work involved in maintaining agent runtimes, while requiring teams to design and oversee agent behavior and permissions.
DigitalOcean Managed Agents combines a durable Harness Runtime with an Action Gateway that provides governed tool access through a managed MCP endpoint. Agents run in hardware-isolated microVMs with capabilities including Chromium and code execution. Lifecycle APIs allow developers to pause, resume, and fork sessions while preserving working state. Developers can deploy supported coding harnesses or their own agents and connect them to more than 16,000 tools without managing the runtime infrastructure. The product entered public preview on September 22, 2026.
Key Features
Durable runtime for persistent agent sessions
Hardware-isolated microVM execution
Managed MCP endpoint for governed tool access
Session APIs for pausing, resuming, and forking while preserving working state
Support for coding harnesses or custom agents, with access to more than 16,000 tools
Real-World Use Cases
Persistent software development agents
Run coding agents in isolated environments that can retain working state across pauses and resumptions, with access to code execution and connected tools.
Tool-connected workflow automation
Deploy agents that interact with external tools through a managed MCP endpoint, subject to configured access governance.
Agent prototyping and evaluation
Test supported coding harnesses or custom agents without maintaining the underlying runtime infrastructure.
Business Opportunities
Managed coding-agent deployment
Problem: Teams building coding agents may need to operate persistent execution environments and connect agents to development tools.
Solution: Use managed runtimes, isolated microVMs, session lifecycle APIs, and governed tool access to host coding agents.
Estimated ROI: Unknown without workload, usage, and pricing data. Potential value comes from reduced runtime maintenance and faster agent experimentation.
Execution: Pilot with a bounded development workflow, define tool permissions and human review requirements, then assess reliability and operating effort before expanding.
Persistent internal workflow agents
Problem: Some multi-step workflows require agents to retain working state and interact with multiple tools over time.
Solution: Deploy agents in persistent sessions and route tool interactions through the managed MCP endpoint.
Estimated ROI: Unknown; depends on workflow volume, automation reliability, integration effort, and platform costs.
Execution: Select a low-risk workflow, map required tools and permissions, test session recovery and agent outputs, and introduce human approval for consequential actions.