# OpenHands
Open-source, model-agnostic platform for building and running autonomous coding agents. Operated by All Hands AI; CEO [[Robert Brennan]], Chief Scientist [[Graham Neubig]] (also CMU LTI). Cloud is hosted at `app.all-hands.dev`; core repo is MIT-licensed and community-maintained under the `OpenHands` GitHub org. Positions itself as "build coding agents, your way" — the engineer-led, self-hostable alternative to closed platforms like [[Devin]] and [[Cursor.com]].
The pitch is *execution*, not suggestion: agents that plan and apply changes end-to-end (open reviewable PRs, fix vulnerabilities, migrate legacy code, triage incidents, refactor at scale) rather than autocompleting code line by line. ~75K GitHub stars, 400+ contributors, 5M+ open-source downloads, $18.8M raised.
## Architecture
Four layers shipped from one repo:
- **Software Agent SDK** — composable Python library; the engine that powers everything else. Pre-built tools for Bash, file editing, web browsing, and [[Model Context Protocol (MCP)]] integration. Also exposes a REST API.
- **CLI** — terminal client that drives the SDK against Claude, GPT, or any other model.
- **Local GUI** — desktop app with a React frontend over the REST API.
- **Agent Server** — production-ready server that runs agents in Docker or Kubernetes, callable from the Python API.
Deployment targets: local dev machine, cloud at scale (thousands of parallel agents), or self-hosted on Kubernetes inside a private VPC. Sandboxed runtimes (Docker / K8s) are non-optional — security is treated as a runtime property, not a promise.
## Manifesto (the opinions worth keeping)
- The future of software development must be written by engineers — not handed down by closed vendors
- A new role is emerging: engineers who orchestrate tens to thousands of cloud agents instead of writing every line themselves
- Local agents are transitional. Real leverage is cloud / VPC parallelism
- Openness is non-negotiable for critical infrastructure; agents must be inspectable
- No model lock-in — swap models per task on economics or quality
- Security by design: isolated sandboxes, scoped credentials, clear blast radius, full auditability
## Integrations
GitHub, GitLab, Bitbucket, Jira, Linear, Slack, CI/CD systems, ticketing tools. Slack/Jira/Linear are part of the hosted cloud tier.
## Licensing
- Core platform (SDK, CLI, GUI, server): MIT
- `enterprise/` directory: source-available; commercial license required for production use beyond a one-month trial
## Sibling projects in the org
`agent-sdk`, `OpenHands-CLI`, `benchmarks`, `openhands-chrome-extension`, `ToM-SWE` (Theory-of-Mind module for the agent).
## Why it matters
OpenHands is the credible open competitor to closed agentic-coding products. For anyone who needs vendor-neutral, auditable, self-hostable agent orchestration — especially in regulated or air-gapped environments — it's the obvious reference implementation. The SDK alone is interesting as a Python harness for [[AI Agents]] that you can wire into your own runtime.
## Orchestration from other harnesses
[[Hermes Agent]] can orchestrate OpenHands agents via an optional first-party skill (announced by [[Teknium]], May 2026):
```sh
hermes update
hermes skills install official/autonomous-ai-agents/openhands
```
Once installed, the skill is auto-discoverable; Hermes finds it when asked to delegate, or you can force-load with `/<agent-name> <prompt>`. Same pattern works for [[Claude Code]], [[Codex CLI]], and OpenCode skills shipped under `official/autonomous-ai-agents/`. This is the cleanest path today to compose OpenHands into a multi-harness workflow without writing glue.
## References
- Website: https://www.openhands.dev/
- Docs: https://docs.openhands.dev
- SDK docs: https://docs.openhands.dev/sdk
- Manifesto: https://www.openhands.dev/manifesto
- Blog: https://www.openhands.dev/blog
- About: https://www.openhands.dev/about
- GitHub: https://github.com/OpenHands/OpenHands
- Hosted cloud: https://app.all-hands.dev
- X: https://x.com/OpenHandsDev
- Hermes Agent OpenHands skill announcement: https://x.com/Teknium/status/2059038964552745378
## Related
- [[AI Agents]]
- [[Hermes Agent]]
- [[Claude Code]]
- [[Devin]]
- [[Cursor.com]]
- [[Model Context Protocol (MCP)]]
- [[Anthropic]]
- [[Teknium]]