OpenHands
by All Hands AI
Open-source AI software engineer that edits, runs, and tests code in a sandbox
Last reviewed 2026-06-19
OpenHands (formerly OpenDevin) is an open-source AI coding agent from All Hands AI. Rather than autocompleting code inline, it operates as an autonomous agent inside a sandboxed environment with its own shell, file system, and web browser: it can edit files, run commands, run tests, browse the web, and iterate on a task, then hand the result back for human review. It ships as an MIT-licensed core (run locally via Docker, CLI, or the GUI), a hosted OpenHands Cloud, an enterprise tier, and a Python Software Agent SDK for building custom agents. OpenHands is one of the most popular open-source coding-agent projects and is positioned as a model-agnostic, no-lock-in alternative to closed agents like Devin. The team recommends frontier models such as Anthropic's Claude (Sonnet) but supports GPT, Gemini, open models, and bring-your-own-key. It can resolve GitHub, GitLab, and Bitbucket issues when triggered by a label or mention, run multiple agents in parallel in the cloud, and is widely used for benchmarking against suites like SWE-bench.
What it can do
Edit, run, and test code end to end
SupervisedWorks inside a sandbox with shell, file system, and browser to edit files, run commands, run tests, and iterate on a task; output is meant to be reviewed and tested by a human.
sourceResolve issues and PR comments from a label or mention
SupervisedTriggered by an `openhands` label or `@openhands` mention on GitHub, GitLab, or Bitbucket, it works the issue and opens a pull request for review.
sourceRun multiple agents in parallel (Cloud)
SupervisedOpenHands Cloud can run several agents concurrently on different tasks with central visibility.
sourceBuild custom agents via the Software Agent SDK
SupervisedA Python SDK exposes the agent loop, tools, and runtime so developers can build and orchestrate their own coding agents.
source
Strengths
- +Fully open source (MIT core) and model-agnostic with no vendor lock-in; self-hostable for privacy
- +Genuinely agentic: edits, runs, and tests code in a sandbox and opens PRs, not just autocomplete
- +Large open-source community plus a mature SDK, REST/WebSocket API, and MCP support
Limitations
- −Open-source build is single-user with no built-in auth/isolation; teams need the Enterprise tier
- −Output must be reviewed and tested, and open-ended tasks can burn many LLM calls
- −Capability is commoditized; benchmark leadership is fleeting and quality depends on the chosen model
Overview
OpenHands (formerly OpenDevin) is an open-source AI coding agent from All Hands AI. It runs inside a sandboxed environment with a shell, file system, and web browser, and is built to do real engineering work rather than autocomplete lines.
What it does
Given a task, OpenHands edits files, runs commands, runs tests, browses the web, and iterates, then hands the result back for human review. It can resolve GitHub, GitLab, and Bitbucket issues when triggered by a label or mention, open pull requests, and (in OpenHands Cloud) run multiple agents in parallel. A Python Software Agent SDK lets developers build and orchestrate their own agents.
Integrations & setup
Runs locally via Docker, a CLI, or a GUI, in OpenHands Cloud, or in a private VPC for enterprises. It connects to GitHub, GitLab, Bitbucket, Slack, Jira, and Linear, consumes MCP servers, and exposes a REST plus WebSocket API. It is model-agnostic with bring-your-own-key; the team recommends Anthropic's Claude Sonnet but supports GPT, Gemini, and open models.
Pricing
The open-source core is free and self-hostable. OpenHands Cloud has a free tier with usage caps (a former paid Pro tier was retired) and a contact-sales Enterprise tier with SSO, RBAC, and VPC deployment.
Best for / not for
Best for developers and teams that want an open, model-agnostic, self-hostable coding agent. Less suited to teams that need a managed, multi-user product out of the box without operating their own infrastructure.
Traction
OpenHands raised a reported $5M seed in September 2024 led by Menlo Ventures, and All Hands AI was named to Anthropic and Menlo's Anthology Fund first cohort. It is among the most-starred open-source coding-agent projects.
Alternatives
Devin is the closest delegated-agent competitor; Cursor and Aider target in-IDE and terminal coding.
What people are saying
We aggregate real LinkedIn discussion into sentiment for the agents people search most. OpenHands isn't tracked yet, want it added? Request tracking.
FAQ
Is OpenHands fully autonomous?+
It runs autonomously inside a sandbox (editing, running, and testing code), but its output is meant to be reviewed and tested by a human, and it typically opens a pull request rather than merging on its own. In practice it operates as a supervised agent.
What was OpenHands called before?+
It was originally OpenDevin, launched in March 2024, and was renamed OpenHands by All Hands AI.
Sources
- OpenHands (official site) · accessed 2026-06-19
- OpenHands documentation · accessed 2026-06-19
- All-Hands-AI/OpenHands (GitHub) · accessed 2026-06-19
- All Hands AI raises seed led by Menlo Ventures (All Hands AI blog) · accessed 2026-06-19
Last reviewed 2026-06-19