OpenHands

Does it fit your conditions?

You need control over an agent's execution environment and can evaluate deployment options.

Setup

Configure tools

Interface

Browser

Access

Provider API key / Product account

Workspace

Independent workspace

Choose managed Cloud or configure the backend, sandbox, repository, and model access.

Reviewed scope: Agent Canvas and Agent Server; managed Cloud is another deployment option.

Coding and office fit

Self-hosting brings infrastructure work. Legacy local interfaces should not be assumed to match Agent Canvas.

Costs and context

Managed usage / provider and infrastructure

For self-hosting, include sandbox compute and provider requests as well as maintenance.

Reduce unnecessary work: Provide a reproducible environment and explicit tests to reduce setup retries.

Measured tokens / task

Unknown — not measured

Measured cost / task

Unknown — not measured

Compare the same deliverable and acceptance criteria, including retries and your review time. Local execution does not establish local model processing.

Try one real task

1. Choose managed Cloud or configure the backend, sandbox, repository, and model access.
2. Start with a small task: Execute and review software-development tasks in an agent environment.
3. Check the result against your original files and acceptance criteria before expanding the workflow.

Before relying on this workflow, check the current plan, device, region, language, file format, and required permissions.

Bring your work with you

Carry source, setup instructions, and task state; recreate the sandbox and verify its tool access.

A reviewed brief can carry context. Native history, internal memory, account permissions, and running tasks are not assumed to transfer.