OpenHands
OpenHands · Open-Source, Self-Hostable Platform for Coding Agents and Automations
OpenHands is an MIT-licensed platform for running software engineering agents, with its own model-agnostic coding agent plus Agent Canvas, a control center that also runs Claude Code, Codex, and Gemini CLI in isolated git worktrees. It suits developers and teams who want coding agents on their own machines or infrastructure, with a hosted cloud and enterprise option when needed.
Best for
Developers and platform teams who want an open-source, bring-your-own-model coding agent they can run and automate on their own machines or VMs
Not ideal for
Non-technical users, teams wanting a fully managed product with no setup, or developers who prefer a polished IDE experience over a beta control center
Who it's for
Developers, open-source maintainers, and engineering teams who want self-hosted, model-agnostic coding agents and automations
Capabilities
- Model-agnostic OpenHands coding agent that works with any configured LLM, including local models via Ollama or vLLM
- Agent Canvas runs multiple agents at once, each isolated in its own git worktree, from one dashboard
- Runs Claude Code, Codex, Gemini CLI, or any Agent Client Protocol (ACP) agent alongside the OpenHands agent, using your existing subscriptions
- Automation service triggers agents from cron schedules, GitHub webhooks, and Slack, with run dashboards and per-run cost tracking
- Backend runs locally, on a self-hosted VM, or in OpenHands Cloud, switchable in the Canvas UI
- Python Software Agent SDK and Agent Server exposing agent execution over REST and WebSocket APIs
- Terminal CLI installable with uv, a binary install script, or Docker
- MCP connections, Agent Skills, and agent profiles that can be scoped to specific MCP servers and secrets
Limitations
- Agent Canvas is labeled beta, the desktop app is an early preview build, and releases often ship several times a week, so interfaces change frequently
- Running the backend directly on your machine gives agents your user account's permissions; the docs recommend a container, sandbox, or VM for isolation
- Results depend on the model you connect; the docs note open-weight and local models can be slow or return malformed responses
- The free Cloud Individual tier is limited to one user and 10 conversations per day
- SAML/SSO, multi-user RBAC, and VPC deployment require the custom-priced Enterprise plan
- Installing Agent Canvas via npm requires Node.js 22.12+ and uv (Docker or the preview desktop app avoid this), which assumes comfort with developer tooling
Use cases
- Running the OpenHands agent with your own API key or a local model to fix bugs and open pull requests
- Comparing Claude Code and Codex on the same task in separate worktrees from one Canvas dashboard
- Scheduling a weekly cron automation that updates dependencies and opens a PR for review
- Triggering an agent from a GitHub webhook or Slack message to triage new issues
- Hosting an always-on agent server on a team VM so agents keep running when laptops are closed
- Building a custom software engineering agent on the Python SDK and Agent Server API
OpenHands pricing
Open Source (Local)
Free
- OpenHands agent with web GUI, terminal UI, and CLI
- Unlimited daily conversations
- Bring your own LLM key
- GitHub, GitLab, and Bitbucket integrations
- Single user, community support
Individual (Cloud)
Free
- Hosted cloud access from desktop and mobile
- 10 daily conversations, 1 user
- Bring your own key or use OpenHands-provided models at cost
- API access plus Jira and Slack integrations
Enterprise
Custom
- SaaS or self-hosted in your VPC
- Enterprise SAML/SSO and multi-user RBAC
- Unlimited users and concurrent conversations
- Priority support
Free tier limits: The Cloud Individual tier allows one user and 10 conversations per day. The local open-source version has no conversation limit.
Note: OpenHands itself is free in the open-source and Individual tiers; model usage is paid to your own provider (BYOK) or bought pay-as-you-go through the OpenHands LLM provider at cost, with no markup.
Technical specs
Available models
OpenHands vs. competitors
OpenHands vs. claude-code
Claude Code is Anthropic's agent tied to Claude models; OpenHands is an open-source, model-agnostic platform whose Agent Canvas can also run Claude Code sessions alongside its own agent.
OpenHands vs. opencode
OpenCode is an open-source terminal coding agent; OpenHands adds an SDK, a visual control center, event-triggered automations, and hosted Cloud and Enterprise options around its agent.
OpenHands vs. orca
Orca is a desktop app for running third-party CLI agents in parallel worktrees; OpenHands also runs third-party agents in worktrees but ships its own agent, an agent server you can host remotely, and scheduled or webhook-triggered automations.
OpenHands vs. devin
Devin is Cognition's proprietary cloud agent, with its agent loop always in Cognition's cloud and usage-metered plans; OpenHands is MIT-licensed, runs on your own machines or VMs with any model, and offers a hosted Cloud only as an option.
Frequently asked questions
What is OpenHands?
OpenHands is an open-source platform for software engineering agents. It includes its own coding agent, a Python SDK and Agent Server, a CLI, and Agent Canvas, a control center for running and automating coding agents in isolated git worktrees.
Is OpenHands open source and free?
Yes. The OpenHands repository is MIT-licensed, and the local open-source version is free with unlimited conversations; you bring your own LLM API key. OpenHands Cloud has a free Individual tier, and Enterprise is custom-priced.
Can OpenHands be self-hosted?
Yes. The agent server can run on your laptop, a dedicated machine, or a cloud VM you control. OpenHands Enterprise adds licensed self-hosting in your own VPC or on-premises with SSO and RBAC.
Which models does OpenHands support?
OpenHands is model-agnostic. Its docs recommend frontier models such as Claude Opus 4.8, GPT-5.5, and Gemini 3.5 Flash, list open-weight options like GLM-5.1 and Kimi-K2.6, and support local models through Ollama or vLLM.
Does OpenHands work with Claude Code or Codex?
Yes. Agent Canvas can run Claude Code, Codex, Gemini CLI, or any agent that supports the Agent Client Protocol, alongside the OpenHands agent, so you keep your existing subscriptions.
Integrations & fit
Alternatives to consider
Anthropic
Developers and teams who want deep codebase reasoning, multi-file changes, and persistent project instructions in a terminal-friendly or IDE-native workflow
Paid
Anomaly
Developers who want a terminal-based AI coding agent with no vendor lock-in, the ability to bring any LLM provider, and undo control over agent file changes
Free
Stably AI
Developers who already use one or more CLI coding agents and want to run and review many of them in parallel without branch conflicts
Free
Cline
Developers and teams who want an open-source, model-agnostic AI coding agent they fully control, in their existing IDE or terminal
Freemium
Cognition
Engineering teams with a steady backlog of well-scoped tickets who want to delegate them asynchronously from Slack, Teams, or their issue tracker and review the resulting PRs
PaidAbout OpenHands
OpenHands has grown from a single open-source coding agent into a stack of layers you can adopt separately. At the bottom is the Software Agent SDK, a Python library and Agent Server that exposes agent execution over REST and WebSocket APIs. On top of it sit a terminal CLI and Agent Canvas, a browser and desktop control center where each conversation runs in its own git worktree and can use the OpenHands agent or any agent that speaks the Agent Client Protocol, including Claude Code, Codex, and Gemini CLI. An automation service adds cron schedules, GitHub webhooks, and Slack triggers, so recurring chores like dependency updates or issue triage run without someone starting them. The backend can run directly on your laptop, on a dedicated machine or cloud VM you control, or inside OpenHands Cloud. The open-source version is free and bring-your-own-key: you pay only your model provider, and any LLM can be configured, from Claude, GPT, and Gemini to open-weight models served locally through Ollama or vLLM. OpenHands Cloud adds hosted sandboxes, a Cloud API, and Jira and Slack integrations on a free single-user tier, and Enterprise adds SAML/SSO, multi-user RBAC, and deployment in your own VPC. The tradeoffs: Agent Canvas is still labeled beta and ships several releases a week, running agents directly on your machine gives them your user permissions unless you add a container or VM boundary, and output quality depends heavily on the model you connect — the docs warn that weaker local models can produce malformed tool calls.
Updates from OpenHands
The v1.23.0 release added a universal macOS DMG with per-architecture bundled runtimes, capping five releases between September 16 and 23, 2026 that added agent profiles scoped to MCP servers, Docker runtime settings, and remote MCP server testing on cloud backends.
OpenHands reported eight Agent Canvas releases in August 2026 with 168 merged pull requests from 47 community contributors, adding Canvas Apps (beta), automation phase indicators and per-run cost tracking, a Commits drawer for review, and child conversations for delegated work.
Are you the founder? Claim this listing →