OpenHands
by All Hands AI
Open-source, self-hosted control center for running coding agents on real engineering work
OpenHands is an MIT-licensed open-source platform for software-engineering agents, built by All Hands AI. It gives developers a GUI, CLI and SDK for running agents that write code, run tests and open pull requests, and it can orchestrate OpenHands, Claude Code, Codex and Gemini agents locally, in Docker or on enterprise infrastructure.
OpenHands launched in spring 2024 as OpenDevin and is now maintained by All Hands AI as one of the most widely adopted open-source alternatives to proprietary coding agents such as Devin, Claude Code and Cursor. The repository carries 84.2k GitHub stars, 10.9k forks and an MIT licence, with the current release at v1.13.0 and 313 open issues. Architecturally it is a scaffold rather than a model: it wraps a sandboxed environment in which an agent can browse the web, edit files across a repository, run shell commands, execute tests and submit pull requests, and it is deliberately model-agnostic so teams supply their own LLM keys. That design explains its benchmark profile, which tracks whichever frontier model it is pointed at — paired with Claude Sonnet 4.5 using extended thinking it reaches 72% on SWE-bench Verified, a fully open self-hostable model configuration scores 32%, and it ranks first on Multi-SWE-Bench across eight languages. During 2026 the product broadened from a single agent into a self-hosted developer control center that also runs Claude Code, Codex and Gemini agents from one dashboard, letting teams switch between local, remote and cloud execution without losing context. Release v1.6.0 in March 2026 added Kubernetes support for enterprise self-hosting, a Planning Mode beta, multi-user RBAC, audit logs and data-residency guarantees for regulated industries. All Hands AI runs an open-core model: the platform is free to self-host, a hosted individual tier is capped at ten daily conversations, and Enterprise adds SAML/SSO, private VPC deployment, a large-codebase SDK and a named customer engineer. The company has raised $23.8 million.
The engineering leader at a regulated or cost-sensitive organisation who needs autonomous coding agents running inside their own VPC or Kubernetes cluster, on their own model keys, with a full audit trail.
Autonomous batch coding work — tests, refactors, boilerplate, issue triage — runs on infrastructure you control at token cost, with no per-seat vendor markup.
At a Glance
- Category
- Developer Tools
- Pricing
- Free, Usage-based, Contact for pricing
- Target Market
- CTOs, VP Engineering, Enterprise Developers, Platform Engineering Leads
- Deployment
- Open-source, Self-hosted, Cloud-first, Hybrid
- Founded
- 2024
Key Features
- ✓Model-agnostic agent scaffold
Bring your own LLM key; agent quality tracks whichever frontier or open-weight model you point it at.
- ✓Sandboxed execution environment
Agents get an isolated filesystem, shell and browser so autonomous runs never touch production systems directly.
- ✓Multi-agent control center
One dashboard runs OpenHands, Claude Code, Codex and Gemini agents across local, remote and cloud environments.
- ✓Kubernetes self-hosting
Shipped in v1.6.0, letting enterprises deploy the whole platform inside their own cluster with data-residency guarantees.
- ✓GUI, CLI and SDK interfaces
Interactive sessions, terminal-native workflows and programmatic embedding into third-party tools are all first-class.
- ✓Enterprise governance
Multi-user RBAC, audit logs and SAML/SSO provide the access controls regulated engineering organisations require.
- ✓Planning Mode
Beta capability that has the agent produce and confirm a plan before executing multi-file changes.
Capabilities
Use Cases
- •Autonomous test and refactor batches
Hand the agent a backlog of refactors or missing tests overnight and review the pull requests it opens by morning.
- •Hybrid coding workflow
Use Cursor or Claude Code interactively while routing high-token batch work to self-hosted OpenHands to cut model spend.
- •Regulated-industry adoption
Deploy on Kubernetes inside the corporate VPC so proprietary source never leaves the network, with audit logs for compliance.
- •GitHub issue triage
Point the agent at an issue backlog to decompose tickets, reproduce bugs and draft candidate fixes for human review.
- •Model cost-benefit evaluation
Benchmark an open-weight model against a frontier model on the same scaffold to quantify what closed models actually buy.
Ideal For
Best For
- ✓Running autonomous coding tasks such as test writing, refactoring and dependency upgrades in a sandboxed environment
- ✓Self-hosting a coding agent inside a private VPC or Kubernetes cluster where source code cannot leave the network
- ✓Teams that want to swap models freely — frontier, open-weight or local — without changing their agent tooling
- ✓Automating GitHub issue decomposition, code review and repository reports from a single dashboard
- ✓Benchmarking agent scaffolds and models against SWE-bench-style tasks before committing to a proprietary vendor
Not Ideal For
- ✗Developers who want an in-editor assistant with inline diffs and real-time pairing — reviewers consistently place Cursor and Claude Code ahead for interactive coding, and describe OpenHands as trading convenience for control
- ✗Teams without DevOps capacity: setup involves Docker-in-Docker, API key configuration and environment tuning, and independent reviews state plainly that it is not a download-and-go experience
- ✗Front-end-heavy work — reviewers report that generation of React components and CSS layouts is noticeably less reliable than backend and API work because the agent cannot see rendered output
Integrations
Deployment
Market Analysis
Pros
- ✓MIT-licensed with 84.2k GitHub stars and 10.9k forks, and contributions from engineers at AMD, Apple, Google, Amazon, Netflix and NVIDIA
- ✓Best-in-class open-source benchmark position: 72% on SWE-bench Verified with Claude Sonnet 4.5, and first place on Multi-SWE-Bench across eight languages
- ✓Genuine enterprise self-hosting with Kubernetes, multi-user RBAC, audit logs and data residency, not a stripped-down community edition
- ✓No per-seat licence at all — cost is model tokens only, which reviewers price at roughly $0.15 to $0.60 for a typical task
Cons
- ✗Setup is heavy: Docker-in-Docker, API key configuration and environment tuning, and reviewers state it is explicitly not a download-and-go experience
- ✗Agents get stuck repeating the same failing approach, and recognising and breaking those loops takes operator experience
- ✗Front-end code generation for React components and CSS layouts is markedly weaker than backend work because the agent cannot see visual output
- ✗A 40-point gap remains between the fully open self-hostable configuration at 32% SWE-bench and the frontier-model configuration at 72%, so true independence costs accuracy
- ✗No published SOC 2 or ISO 27001 certification, and the free hosted tier is capped at ten conversations per day
Pricing
Open Source (Local)
$0
- ✓Runs locally, single user
- ✓Web GUI, terminal UI and CLI
- ✓Git integrations
- ✓Model agnostic
- ✓Community support
Individual (SaaS)
$0
- ✓Hosted cloud access on desktop and mobile
- ✓Bring your own key, or use OpenHands providers at cost with no markup
- ✓Jira and Slack integrations
- ✓API support
- ✓Capped at 10 daily conversations
Enterprise
Contact for pricing
- ✓SaaS or self-hosted in a private VPC
- ✓Enterprise SAML/SSO
- ✓Unlimited users and concurrent conversations
- ✓Large Codebase SDK
- ✓Priority support, shared Slack channel and named customer engineer
The software itself is free under an MIT licence, so what you actually pay for is model tokens. OpenHands resells its own LLM provider at cost with explicitly no markup, and independent reviewers measured typical tasks at 50k-200k tokens, roughly $0.15 to $0.60 each on a frontier model. The free hosted tier is hard-capped at ten conversations per day per user. SAML/SSO, private VPC deployment, unlimited concurrency, the Large Codebase SDK and a named customer engineer are Enterprise-only with unpublished pricing.
Security & Compliance
Connect
Sources
This page was written from 6 sources, 4 on domains other than openhands.dev.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Confident AI
Hosted LLM evaluation, observability and red teaming built on the open-source DeepEval framework
Respan
Observability, evals and an LLM gateway for AI agents in one control plane
Meta Muse Code
Meta's terminal coding agent for large repositories, with persistent background agents and the most aggressive token pricing in the category
Niteshift
The full-stack cloud for coding agents — real environments, verified pull requests