🩺 Vitals
- 🟢 Last active: 2026-08-21
- 📦 Latest release: v1.14.0 (2026-08-17)
- 🐞 Open issues: 517
- 🌟 Stars: 84.7k
What do these metrics mean?
- Last active: when code was last pushed, as of our last check. The dot is green when that was recent, grey otherwise. A long gap can mean a tool is finished and stable, not only unmaintained.
- Latest release: the most recent tagged, packaged version the maintainers published. Not every healthy project tags releases.
- Open issues: unresolved reports and requests. A high number is normal for a popular project and is not a warning on its own.
- Stars: how many people bookmarked the project on its forge. A rough popularity signal, not a measure of quality.
🏗️ Profile
- Official: openhands.dev
- Source: github.com/OpenHands/OpenHands
- License: MIT (Core) / Polyform Free Trial (Enterprise)
- Deployment: Docker | Kubernetes
- Data Model: Local File Mounts / Docker Volumes
- Jurisdiction: United States 🇺🇸 (All Hands AI, Inc., VC-backed)
- Compliance (SaaS): N/A
- Compliance (Self-Hosted): Self-Hosted (User Managed)
- Complexity: High (4/5) - Sandboxed Docker/K8s runtime for untrusted code
- Maintenance: Medium (3/5) - Container image and model config upkeep
- Enterprise Ready: Medium (3/5) - SSO, RBAC & audit logs gated behind commercial tier
1. The Executive Summary
What is it? OpenHands is an autonomous AI software-development agent: hand it a task or a GitHub issue and it plans, writes, runs, and debugs code inside a sandboxed Docker runtime, iterating until tests pass. It is model-agnostic (bring your own LLM key, including local models), and because it operates on local file mounts, the source code it generates never leaves your machine. It leads the open field on SWE-bench-style task-resolution benchmarks. The catch is the licensing seam: the core is genuinely MIT, but the controls an organization needs to run it at scale are not.
The Strategic Verdict:
- 🔴 For teams needing governed multi-seat rollout out of the box: Caution. SSO, RBAC, audit logs, LLM budgeting, and prompt-injection guardrails are gated behind the commercial Polyform-licensed enterprise tier. The MIT core is built around the individual developer.
- 🟢 For sovereignty-focused engineering teams running their own models: Strong Buy. Code execution and source stay inside your perimeter, the model choice is yours, and there is no per-task metering like the proprietary incumbents charge.
2. The "Hidden" Costs (TCO Analysis)
| Cost Component | Devin (SaaS) | OpenHands (Self-Hosted) |
|---|---|---|
| Agent Access | Monthly subscription plus metered per-task compute (ACU) consumption | Free MIT core; you pay only your own LLM API tokens or local model compute |
| Code & Data Residency | Source and execution run in the vendor's cloud | Sandboxed runtime on your own Docker / Kubernetes; code stays on local mounts |
| Enterprise Controls | Bundled into enterprise plans | SSO, RBAC, audit logs, and budgeting gated behind the commercial OpenHands enterprise tier |
3. The "Day 2" Reality Check
🚀 Deployment & Operations
- Installation: Docker is the core dependency: the agent spins up a sandboxed container to execute LLM-generated code, run from the CLI, headless, or a web UI. Point it at your own LLM API key or a local model. The enterprise tier adds a Kubernetes-native runtime for concurrent, isolated agent sandboxes.
- Scalability: Single-developer use runs locally on one machine; org-scale concurrency depends on the enterprise Kubernetes runtime to schedule and isolate agent sandboxes safely.
🛡️ Security & Governance (Risk Assessment)
- Jurisdiction & the CLOUD Act: All Hands AI, Inc. is US-domiciled, so any managed or SaaS use carries CLOUD Act exposure. Self-hosting the MIT core removes vendor-held data from scope entirely: there is nothing in the vendor's hands to subpoena.
- The Compliance Shift & Untrusted Code Execution: OpenHands executes LLM-generated code, so the security burden is unusually concrete. You own the Docker / Kubernetes sandbox hardening, container isolation, data persistence, and RBAC. A misconfigured sandbox is a direct remote-code-execution risk, not merely a compliance checkbox.
- The Open-Core Trap: The core is genuinely MIT, but the controls an enterprise actually needs (SSO, RBAC, audit logs, LLM budgeting, prompt-injection guardrails) sit under a commercial Polyform Free Trial license. Sovereignty is real for an individual developer; at organizational scale you either build those controls yourself or pay the enterprise tax.
4. Market Landscape
🏢 Proprietary Incumbents
- Devin: The benchmark-setting autonomous software engineer, but fully cloud-hosted with subscription plus per-task metering and no self-hosting path: source and execution live in the vendor's cloud.
- Cursor: An AI-native IDE with an agent mode. Faster for in-editor, human-in-the-loop edits, but code and context flow through its cloud rather than a sandbox you control.
🤝 Open Source Ecosystem
- Cline: IDE-embedded agentic coder (bring-your-own-key). More interactive and human-in-the-loop than OpenHands' fully autonomous, sandboxed task resolution.
- Goose: Block's local, model-agnostic agent for engineering tasks. Shares the sovereignty stance but is less specialized around sandboxed, full-issue resolution.