Read and edit code
Explore a repository, understand how it fits together, and make multi-file changes rather than isolated snippets.
Core capabilities
The things an agent can actually do while working on a task.
Run commands
Execute shell commands in a sandboxed workspace — install dependencies, run builds, run test suites, and read the output.
Interactive terminal
Drive interactive tools such as a Python REPL or IPython, not just one-shot commands.
Browse the web
Open pages to check documentation or verify a running app. Browser sessions can be recorded and replayed.
Use external tools
Connect to Model Context Protocol (MCP) servers and custom tools, so agents can reach your own systems and APIs.
Delegate to sub-agents
Hand off scoped work to sub-agents that run and report back, which keeps large tasks manageable.
Keep context
Persistent memory, context condensing, and reusable skills help an agent stay useful across long or repeated tasks.
Stay under control
Action confirmation, pause and resume, conversation forking, and budget or usage limits keep a human in the loop.
Typical tasks people hand to it
Ordinary engineering work that is well-specified but tedious.
- Fix a failing test
- Resolve a GitHub issue
- Review a pull request
- Bump and patch dependencies
- Write missing tests
- Refactor a module
- Add a small feature
- Write or update docs
- Migrate an API
- Triage and label issues
- Track down a bug
- Generate a recurring report
How a task runs
Agents work in a loop rather than producing a single answer.
- You describe the goal In plain language — from a chat message, a terminal prompt, an issue, or a scheduled trigger.
- The agent explores It reads the relevant files and runs commands to learn how the project is built and tested.
- It makes changes and checks them Edit, run, read the result, correct course. Failures become feedback for the next step.
- It reports back You get the diff, the commands it ran, and — where integrations are set up — a branch or pull request to review.
Ways to use it
Same agents, different front doors.
| Surface | What it is |
|---|---|
| Agent Canvas | Open-source browser client and control center for agent conversations and automations. Connects to one or more Agent Server backends — local, self-hosted, or cloud. |
| CLI | Terminal interface built on the Software Agent SDK, for working from a shell. |
| Software Agent SDK | Composable Python library for building your own agents, tools, and workspaces. |
| IDE editors | VS Code, JetBrains, Zed, and other clients via the Agent Client Protocol (ACP). |
| OpenHands Cloud | Managed service, so you don't operate your own backend and sandbox infrastructure. |
| Enterprise | Licensed self-hosting or managed deployment, with commercial features and support. |
Integrations and automation
Agents can be triggered by your existing tools instead of only by a chat box.
Code hosting
GitHub, GitLab, and Bitbucket — including PR review, reviewer assignment, and repository monitoring. Azure DevOps is available for Enterprise.
Chat and project tracking
Slack channel monitoring and Jira Cloud, so work can start from a message or a ticket.
Scheduled and event-driven runs
An Automation Server runs jobs on a schedule or in response to webhook events — nightly checks, recurring reports, reactions to a new issue.
Bring your own model
OpenHands can connect to any LLM supported by LiteLLM, so you choose the provider and keep the keys. Agentic coding is demanding, though, so a capable model makes a large difference to results.
- Anthropic
- OpenAI
- Azure
- AWS Bedrock
- Google Gemini / Vertex
- Groq
- OpenRouter
- LiteLLM Proxy
- Ollama
- vLLM
- LM Studio
- SGLang
Local models are supported, so an agent can run entirely on your own hardware.
What to keep in mind
An agent that can edit files and run commands is powerful in both directions. Give it a sandboxed workspace rather than unrestricted access to your machine, review its diffs before merging, and treat its test runs as evidence to check rather than a guarantee. Results also depend heavily on the model you connect and how clearly the task is described.
Where to go next
The documentation at docs.openhands.dev covers installation, configuration, and integrations. Source code lives across the OpenHands repositories on GitHub, and there is a community Slack. Licensing is per repository, so check the specific project you plan to use.