AI coding agents are becoming a more established part of software development, taking on tasks that involve navigating repositories, modifying code, running tests, and working across multiple development tools. As developers expand how they use agents, having access to the right models and execution infrastructure becomes increasingly important.

Developers need the flexibility to select models suited to different tasks, run agents in environments that support their workloads, and manage those agents without introducing unnecessary operational complexity.

We’re expanding that flexibility through our integration with DigitalOcean.

With OpenHands v1.27.0, developers can now create and manage DigitalOcean-hosted OpenHands agents directly from Agent Canvas. The new integration builds on OpenHands’ support for DigitalOcean Serverless Inference, giving developers the flexibility to use DigitalOcean for agent execution, model inference, or both.

By bringing DigitalOcean’s managed agent infrastructure into Agent Canvas, we’re expanding where developers can run OpenHands agents while maintaining the same open, model-agnostic development experience.

Making cloud-hosted agent development more accessible

AI coding agents need execution environments that support the work they’re being asked to perform. Investigating a bug, implementing a feature, or reviewing a pull request can involve reading code, executing commands, running tests, and interacting with external tools.

For developers working locally, OpenHands provides an environment where agents can perform these tasks directly. Cloud-hosted execution extends those capabilities to remote infrastructure, giving developers additional flexibility in how they allocate compute resources and manage agent workloads.

DigitalOcean Harness Runtime provides managed, isolated execution environments for AI agents. Each OpenHands session runs in a dedicated, isolated microVM sandbox with its own disk, terminal, and browser tools, giving agents the resources to execute code and complete development tasks without developers having to provision and manage the underlying infrastructure.

With the new integration, developers can create DigitalOcean-hosted agents, connect to existing sessions, and manage their work directly from Agent Canvas. The workspace remains consistent whether agents run locally, on DigitalOcean, or in other supported environments.

This is particularly useful as developers begin incorporating agents into more complex workflows, where tasks may require dedicated compute resources or benefit from execution environments independent of their local machines.

For developers already building on DigitalOcean, the integration provides a straightforward way to bring OpenHands into their existing cloud infrastructure. For OpenHands users, it adds another managed execution option within the same workspace.

Connecting DigitalOcean to OpenHands Agent Canvas

The integration introduces DigitalOcean as a supported backend in Agent Canvas, allowing developers to create and manage cloud-hosted OpenHands agents alongside their existing local and remote environments.

Developers can authenticate with DigitalOcean and access their existing agents or provision new cloud-hosted instances through the Agent Canvas interface.

Once connected, developers can:

• Create cloud-hosted agents: Provision OpenHands agents on DigitalOcean, including configurable sandbox sizes where supported.

• Connect to existing agents: Access DigitalOcean-hosted agents and their sessions directly from Agent Canvas.

• Manage agent sessions: Resume paused sessions, connect to running sessions, and pause or stop sessions from the workspace.

• Work across environments: Manage DigitalOcean-hosted agents alongside other supported local and remote backends.

Agent Canvas connects to DigitalOcean-hosted OpenHands agent servers through a port-forwarded connection, allowing developers to interact with remote agents using the same tools and interface they use for local development.

Sessions can be paused when they’re not in use, preserving files, installed tools, and conversations while stopping compute charges. Developers can resume their work directly from Agent Canvas, without having to recreate their execution environment or start a new conversation. DigitalOcean also automatically pauses idle sessions to help manage compute costs.

These capabilities make it easier to work on development tasks across multiple sessions while maintaining control over the underlying compute resources.

Choose the models that power your agents

Infrastructure is one part of the agent development experience. Model selection is another.

Different software engineering tasks place different demands on models. A lightweight model might be appropriate for straightforward code changes or routine maintenance, while more complex debugging and reasoning tasks may benefit from models with greater capabilities.

OpenHands is designed to let developers choose the models that best fit their requirements rather than tying agent execution to a single provider.

OpenHands supports DigitalOcean Serverless Inference, allowing developers to access supported models through a DigitalOcean model access key. Developers can select DigitalOcean as their model provider in Agent Canvas, configure reusable LLM profiles, and choose from available models without manually configuring an inference endpoint.

DigitalOcean Serverless Inference works independently of the execution environment, so developers can use DigitalOcean-hosted models whether their agents run locally, on DigitalOcean Managed Agents, or on another supported backend.

This gives developers a centralized way to configure model access and maintain reusable settings for their agent workflows.

The model-provider connection and Managed Agents integration support two distinct parts of the development experience. Model profiles determine which models an OpenHands agent uses, while execution backends determine where that agent runs.

Developers can use DigitalOcean for model inference, agent execution, or both, while retaining the flexibility to configure their OpenHands environment around their specific requirements.

Supporting an open, flexible approach to AI-driven development

As organizations incorporate AI agents into more of the software development lifecycle, their infrastructure requirements will vary. Some teams will prioritize the simplicity of managed cloud services, while others need self-hosted or air-gapped deployments to meet security, compliance, or operational requirements.

OpenHands is designed to support these different approaches. By separating the agent experience from the underlying models and execution environments, developers can choose the infrastructure that works for their organization without changing how they interact with agents.

Our DigitalOcean integration builds on that architecture, adding managed agent execution and model inference options to an already flexible platform.

For developers, it means more choice in how they configure and run agents. For organizations, it provides another way to adopt agent-driven development without committing to a single model provider or deployment environment.

Get started with OpenHands on DigitalOcean

DigitalOcean Managed Agents support is available in OpenHands Agent Canvas v1.27.0 and later. DigitalOcean Managed Agents is currently in public preview.

To get started, open Agent Canvas and navigate to Manage Backends → Add Backend → Providers and select DigitalOcean. Connect using your DigitalOcean personal access token, then create a new OpenHands agent or connect to an existing one.

Once connected, configure an LLM profile using your preferred model provider or DigitalOcean Serverless Inference. You can then start a conversation and begin working with your agent on DigitalOcean’s managed infrastructure.

Ready to get started? Follow the ⁠OpenHands DigitalOcean setup guide or the ⁠DigitalOcean deployment guide to configure your first agent.

Learn more about ⁠DigitalOcean Serverless Inference in OpenHands.