OpenHands

OpenHands is an open-source coding agent and the tooling around it. You give it a task in plain language and it works in a workspace, reading and editing files, running shell commands and opening pull requests. It is aimed at developers who want an agent they can run on their own hardware with any model, and who want agents to keep working on schedules or events, not only while they watch.

There are several ways in. The CLI (installed with uv or as a binary) runs the agent in your terminal, has a headless mode for scripts and CI, and plugs into Zed, JetBrains IDEs and VS Code through the Agent Client Protocol. Agent Canvas is a browser-based control center: it manages conversations, files, terminals and model settings, and connects to one or more backends, such as your laptop, a Docker container per conversation, a VM, Kubernetes or OpenHands Cloud. Besides its own agent, Canvas can run Claude Code, Codex or Gemini CLI, and it has automations triggered by a schedule, GitHub events or webhooks, with prebuilt ones for issue-to-PR, PR review and Slack monitoring. A Python SDK underlies both.

It works with any model LiteLLM supports, including local models through Ollama, LM Studio or vLLM, so you pay your provider directly. OpenHands Cloud is the hosted version: you sign in with GitHub, GitLab or Bitbucket, and it adds hosted sandboxes, Slack and Jira integrations and a cloud API. Its Individual tier is free, with your own key or OpenHands' model provider at cost. Enterprise adds multi-user access, SSO and self-hosting in your VPC.

The open-source version is designed for a single user; the docs say it has no built-in authentication or isolation for shared, multi-tenant use. How safe it is depends on the backend: a local process backend can touch anything your user account can. Result quality depends on the model you connect, and Agent Canvas is labelled beta.