OpenHands hosting

A software-engineering agent that writes, runs and debugs code.

Agents & RAGMITfrom $18.99/mo

In short

OpenHands hosting at AIHostBox means a dedicated, single-tenant OpenHands instance on our infrastructure: we install it, pin it to a reviewed release, put it behind TLS on your domain, back it up nightly and keep it patched. Plans start at $18.99 per month and the admin account, the data and any API keys are yours.

Sandboxed workspaceTerminalBrowserGitBring your own model

OpenHands, formerly OpenDevin, is an MIT-licensed coding agent developed in the open by the All Hands AI team and a large contributor base. Unlike an editor plug-in that suggests lines, it works a task end to end: it reads the repository, edits files, runs commands and tests in a terminal, browses documentation when it needs to, and reports back with a diff. You watch in a web UI on port 3000, or drive it from its CLI and headless mode.

Each conversation runs in its own Docker sandbox started from an agent-server image, so the agent can install packages and execute code without touching the host. That is exactly why it wants a server: a laptop running several sandboxes plus a browser gets hot and noisy, and a task that takes forty minutes should not stop when the lid closes. On a hosted instance you start the job, leave, and review the result later.

OpenHands hosting plans

ProRecommended minimum

8 GB RAM · 4 vCPU · NVMe

$18.99/mo

Order

MaxLarge document sets

12 GB RAM · 6 vCPU · NVMe

$27.99/mo

Order

Every plan is a dedicated single-tenant instance with NVMe storage, nightly off-box backups, TLS and managed patching. Quarterly, semi-annual and annual terms are discounted at checkout. Model tokens are not included — you bring your own key.

How we run OpenHands

The OpenHands app container on port 3000 with access to the instance’s Docker engine, which it uses to start a separate agent-server sandbox container for each conversation, plus a persistent state directory.

What we configure for you

Agent-server image repository and tag pinned to match the app release
LLM_MODEL, LLM_API_KEY and, if needed, LLM_BASE_URL for your provider or gateway
State directory on persistent storage for settings, history and secrets
Access limited to your login in front of the web UI, since the agent can run arbitrary code
GitHub or GitLab token scoped to the repositories you choose

Which plan fits

Pro (8 GB, 4 vCPU) runs the app plus one or two active sandboxes on a typical web or Python repository. Max (12 GB, 6 vCPU) is for larger repositories with heavy builds or test suites, or several conversations running at the same time.

The honest limits

The quality of the work is set by the model you pay for, and CPU-hosted local models are too slow and unreliable at tool use for real tasks. Frontier-model token costs on long tasks are yours and can add up.

What people run OpenHands for

Background refactors

Hand over a mechanical change and review the diff afterwards.

Issue triage

Let it reproduce a bug report before a human picks it up.

Test coverage catch-up

Ask it to write tests for an untested module and run them until they pass.

Dependency upgrades

Bump a library, fix what breaks and hand back a branch for review.

Licence and version policy

Item	Position
Upstream licence	MIT
Sold as managed hosting	Yes
Tenancy	Single tenant — one customer per instance
Patching	Critical advisories within 24 hours, high severity within 72

Our full position on licences, including the applications we decline to sell, is on the licensing page.

Reading on OpenHands

ComparisonOpenClaw alternatives Guide · AgentsAI agent hosting: where an agent should actually run

Works well with OpenHands

More Agents & RAG we host

Questions

OpenHands hosting: your questions

Does it run code on my instance?

Yes, inside a sandbox on your own single-tenant instance. Nothing executes on shared infrastructure and no other customer’s workload is in that sandbox.

Which models work best?

OpenHands depends on strong tool use. Its own index puts frontier models from Anthropic, OpenAI and Google at the top, with open-weight models such as GLM, Kimi and MiniMax via OpenRouter behind them. You set LLM_MODEL and LLM_API_KEY, or LLM_BASE_URL for an OpenAI-compatible gateway.

Can I use a local model on CPU?

Technically yes, through an OpenAI-compatible endpoint, but it is not practical on CPU. The project warns that local and small models give inconsistent tool calls and malformed JSON, and at 5-15 tokens/s a multi-step coding task crawls. Use a hosted model key for real work.

How does it reach my repositories?

Give it a GitHub or GitLab token in settings and it can clone, create a branch and push. We suggest a fine-grained token limited to the repositories you want it to touch, so the agent cannot reach anything else in your organisation.

What is kept between sessions?

Settings, conversation history and stored secrets live in the OpenHands state directory, which is on persistent storage and in the nightly backup. Sandbox containers are disposable; push work to a branch rather than relying on files left inside a finished sandbox.

Run OpenHands without running a server

A dedicated OpenHands instance, deployed and maintained by us. Month to month, cancel whenever.

Browse the catalogue Talk to a human