OpenHands is an open-source platform for developing and deploying AI agents that autonomously perform end-to-end software development tasks within sandboxed environments. Agents can write code, run commands, browse the web, and call APIs.
OpenHands: Autonomous AI Software Engineer
OpenHands enables AI agents to perform end-to-end software development tasks. This includes writing, modifying, and refactoring code, debugging, and resolving GitHub issues. Agents can run shell commands, browse the web for information, interact with APIs, and manage files within a secure, sandboxed Docker environment. This setup ensures that any changes are isolated and don’t affect the host system. The platform employs an "agentic loop" where it receives a task, plans its approach, writes code, executes it, observes the output, and iterates without constant human intervention. An SDK is also available for building and orchestrating custom agents and workflows.
Developers Automating Repetitive Coding Tasks
OpenHands is designed for software developers, engineering teams, and organizations looking to automate repetitive or low-creativity engineering tasks. Specific use cases include:
- Automating code reviews and bug fixes.
- Generating and maintaining tests.
- Automating documentation and release notes.
- Refactoring legacy codebases (e.g., COBOL to Java).
- Vulnerability remediation and incident triage.
- Dependency upgrades.
- Creating new features or applications from specifications.
- Streamlining CI/CD pipelines.
Its self-hosted and sandboxed nature makes it particularly useful for teams with strict data security and compliance needs.
Docker Sandbox, Multi-LLM, and Terminal/Browser Access
OpenHands utilizes an event-stream architecture with key components like an Agent (for the reasoning-action loop), Conversation, an LLM (provider-agnostic interface), a Tool System, and a Workspace (for sandboxed execution). It supports a variety of LLMs, including Claude 3.5/4.5, GPT-4o/5, Gemini 3.0, and local models like Llama 3 via Ollama. All agent sessions run within isolated Docker containers, providing a secure execution environment.
The platform integrates with:
- Git: GitHub, GitLab, Bitbucket for issue-to-PR workflows.
- Collaboration Tools: Jira, Slack, Linear.
- IDEs: Experimental integration via Agent Client Protocol (ACP) for Zed, Toad, VS Code (community extension), and JetBrains IDEs.
- CI/CD: Integrates with CI/CD pipelines, notably GitHub Actions, through its SDK.
OpenHands supports a substantial context window of 200k tokens. It has achieved competitive performance on benchmarks like SWE-Bench, with a 26% solve rate, and claims to autonomously resolve 87% of bug tickets on the same day. It can handle projects with up to 28,000 files.
Free and Open-Source, API Costs Depend on Provider
The core OpenHands platform is open-source and free to use under the MIT license for local self-hosting. Users are responsible for the costs of the underlying Large Language Models (LLMs) they choose to integrate.
| Plan | Price | Key Details |
|---|---|---|
| Free (Local) | Free | OpenHands Agent, web GUI, Terminal UI, CLI, Git integrations, community support. Users bring their own LLM keys. |
| Free (SaaS Individual) | Free | Hosted cloud access, API support, Jira/Slack integrations. Users bring their own LLM key or use OpenHands’ providers at cost. |
| Growth | $500/month | Includes Individual plan features plus unlimited users, shared projects, multi-user RBAC, centralized billing, ticket support. |
| Enterprise | Custom | Self-hosted or private VPC deployments, full control, enterprise SAML/SSO, dedicated support. |
Requires Docker, Agent May Loop on Complex Tasks
OpenHands has practical setup considerations for new users:
- Requires significant setup effort, especially for self-hosting, with Docker knowledge recommended.
- Performance depends heavily on the connected LLM, and agents can sometimes get stuck in repetitive loops.
- Still an early-stage project with evolving APIs and occasional instability in complex workflows.