What You Can Buildpublic
Last verified 21 Sep 2026
DigitalOcean Harness Runtime provides managed, hardware-isolated microVM sandboxes with built-in tools such as Chromium to run harnesses and execute arbitrary code. Rich lifecycle APIs preserve conversational history and working state across sessions, letting you pause, resume, and fork work to control costs and adapt to the nonlinear nature of agentic workflows. Scale complete agents such as Claude Code or use sandboxes independently for code execution, all through the same service. See What You Can Build for example use cases.
Managed Agents runs AI agents on DigitalOcean infrastructure. You bring the agent (Claude Code, Codex, OpenCode, LangGraph, Hermes, or a custom sandbox template), and Managed Agents runs it in an isolated sandbox with session lifecycle controls.
Managed Agents has two products:
- Harness Runtime starts and persists agent sessions in isolated sandboxes.
- Action Gateway gives those agents governed access to tools and APIs through a managed MCP endpoint.
Worked Examples
Each example under Harness Runtime Examples is a complete environment spec with the commands to run it and the fields to change to make it yours.
| Example | What it does |
|---|---|
| Refactor a Repository | Run a long refactor in a sandbox, disconnect, and resume the same workspace later. |
| Review Pull Requests | Fire an unattended first-pass review from a GitHub webhook and deliver it to Slack. |
| Draft Marketing Content | Research a topic weekly, draft copy in your brand voice, and commit it for review. |
| Review Documents | Check contracts against a clause checklist in a locked-down environment. |
| Triage Incidents | Gather context from your observability stack when an alert fires. |
| Serve Tenants From One Environment | Start an isolated session per customer from one saved configuration. |
Capabilities the Examples Rely On
| Capability | What it gives you | Guide |
|---|---|---|
| Session lifecycle | Start work, disconnect, and resume later from another device with workspace state and history preserved. | Manage Sessions |
| Checkpoint, fork, and rollback | Snapshot a session, branch it several ways to compare results, or undo a bad change. | Checkpoint, Fork, and Roll Back |
| Triggers | Run agents unattended on a schedule or from an inbound webhook, and deliver output to email or Slack. | Run Agents With Triggers |
| Permission policies | Decide per tool whether the agent acts freely, asks first, or is blocked. | Configure Permissions |
| Governed tool access | Reach GitHub, Slack, web search, and thousands of other tools with credentials brokered outside the model context. | Connect Action Gateway |
| Environment Configs | Save a configuration once and start any number of sessions from it. | Use Environment Configs |
| Isolated execution | Run code, shell commands, and tests in a microVM that shares no host with another session. | Sandboxes |
| Custom agents | Package your own agent as an OCI image and create sessions from it. | Use a Custom Template |
How Harness Runtime and Action Gateway Work Together
Harness Runtime provisions a session in an isolated microVM, manages pause and resume, supports checkpoints and forks, runs scheduled and webhook triggers, applies per-session permission policies, and exposes session observability.
Action Gateway handles tool discovery, credential brokering, schema validation, retries, output normalization, and tracing for tools the agent calls.
Sessions consume compute while they run work. Pausing a session stops active compute charges while preserving workspace state. For billing details, see Harness Runtime Pricing and Action Gateway Pricing.
Who It Is For
- Engineering teams running their own agents with Claude Code, Codex, OpenCode, LangGraph, or Hermes who want durable cloud execution instead of a laptop or self-managed Droplet.
- Teams building automated agent workflows that need scheduled or event-driven runs with governed tool access.
- Companies whose product is an agent and who want the sandbox, lifecycle, and tool layer managed for them.
- Individual developers who start a session from a terminal or partner UI, check progress from another device, and approve sensitive actions when the permission policy requires it.