OpenHands

What it does

Writes code

YesCore purpose: the agent authors code from a natural-language task across GUI, CLI and SDK surfaces, including scripted headless runs.Build AI agents that write software. A clean, modular SDK with production-ready tools.docs.openhands.dev/llms.txt ↗

Debugs and fixes

YesDocumented for CI failure remediation and incident work: it investigates failures, finds likely causes and implements the fix.Investigate failed builds and tests, identify likely causes, and propose or implement fixes.openhands.dev/services ↗

Writes tests

YesTest generation and coverage maintenance is named as a supported workflow, not only code authoring, on the services page.Generate tests, expand coverage, and keep test suites aligned as the codebase changes.openhands.dev/services ↗

Reviews code

YesA dedicated PR review workflow reads someone else's diff and posts inline comments on the pull request within minutes.The OpenHands PR Review workflow is a GitHub Actions workflow that: Triggers automatically when PRs are opened or when you request a reviewdocs.openhands.dev/openhands/usage/use-… ↗

Whole-codebase work

PartlyRepository-wide migrations are documented as prompts the agent handles, but the Large Codebase SDK that targets big repos is an Enterprise-only line on the pricing table.openhands.dev/pricing ↗

Ships it

PartlyIt ends at a pull request rather than a deploy: the documented automation loop opens a PR and leaves the merge and release to you.When the trigger fires, the agent runs in an isolated cloud container, does the work, and opens a pull request. You review the output.openhands.dev/enterprise ↗

Done-for-you service

PartlySoftware you run yourself, but the vendor also sells advisory and implementation engagements that deploy agents alongside your team.We work alongside your team to design and deploy production-ready agents that are integrated with your environment and workflows, on an OpenHands…openhands.dev/services ↗

How much it does unattended

Runs autonomously

YesEvent-triggered runs complete the task with nobody in the loop, and headless CLI mode forcibly approves every action.Headless mode always runs in `always-approve` mode. The agent will execute all actions without any confirmation. This cannot be changeddocs.openhands.dev/openhands/usage/cli/… ↗

Multi-agent

PartlyA parent agent can delegate to sub-agents, but the documented TaskToolSet runs them sequentially and blocks, so it is delegation rather than parallel coordination.The TaskToolSet lets a parent agent launch sub-agents that handle complex, multi-step tasks autonomously. Each sub-agent runs synchronously - the…docs.openhands.dev/sdk/guides/task-tool… ↗

Agent permissions

YesThree confirmation policies ship in the SDK, from approving every action to only risky ones judged by a security analyzer.AlwaysConfirm() - Require approval for all actions; NeverConfirm() - Execute all actions without approval; ConfirmRisky() - Only require approval for…docs.openhands.dev/sdk/guides/security.… ↗

Which models it runs on

Claude

YesClaude Sonnet, Opus and Haiku are all sold through the vendor's own LLM proxy, and can equally be reached with your own Anthropic key.claude-sonnet-4-5-20250929 | $3.00 | $0.30 | $15.00 | 200,000 | 64,000docs.openhands.dev/openhands/usage/llms… ↗

GPT

YesGPT-5 and GPT-5 Codex appear in the vendor's own model price list, and Azure OpenAI is a documented provider route too.gpt-5-codex | $1.25 | $0.125 | $10.00 | 272,000 | 128,000docs.openhands.dev/openhands/usage/llms… ↗

Open models

YesOpen-weight models are supported first-class: a recommended local model plus documented LM Studio, Ollama, vLLM and SGLang serving paths.We now recommend Qwen3.6-35B-A3B as the first local model to try with OpenHands. It is an open-weight MoE model built for agentic codingdocs.openhands.dev/openhands/usage/llms… ↗

Model choice

YesYou choose the model on every tier; the vendor lists model agnosticism as a feature of the free local plan itself.openhands.dev/pricing ↗

Where you use it

In your editor

PartlyZed, JetBrains and a community VS Code extension connect over ACP, but the vendor itself labels the IDE integration experimental.IDE integration via ACP is experimental and may have limitations.docs.openhands.dev/openhands/usage/cli/… ↗

On the command line

YesA terminal CLI is included on every plan including the free local one, with headless and JSON output modes for scripting.OpenHands CLIopenhands.dev/pricing ↗

In your pipeline

YesShips as GitHub Actions plugins for PR review and QA, and headless mode is documented specifically for CI/CD pipelines.Headless mode runs OpenHands without the interactive terminal UI, making it ideal for: CI/CD pipelinesdocs.openhands.dev/openhands/usage/cli/… ↗

In a browser

YesA web GUI is on every tier, hosted on OpenHands Cloud for the SaaS plans and run locally on the open-source one.Hosted cloud access from desktop and mobileopenhands.dev/pricing ↗

Whose machine it runs on

Self-hosted

YesTwo self-host routes: the free open-source stack on your own machine, and Enterprise deployed into your own Kubernetes VPC.Deploy on your own Kubernetes infrastructure. Your code never leaves your environment. No shared tenancy.openhands.dev/enterprise ↗

Open source

PartlyThe core is MIT-licensed open source, but the Enterprise control plane is described only as source-available under a commercial licence.OpenHands Open Source is a full, MIT licensed version of OpenHands that is designed for a single user and can run locally on your machine with your…openhands.dev/pricing ↗

What it costs to run

How it meters

YesNot seats: you pay for model tokens, either with your own key or through the vendor's proxy billed per million tokens at provider rates.Pricing follows official API provider rates. Below are the current pricing details for OpenHands models:docs.openhands.dev/openhands/usage/llms… ↗

Free tier

YesTwo standing free plans with no sales call: the local open-source build, and a free hosted Individual tier on the cloud.Bring your own key or use our providers at-costopenhands.dev/pricing ↗

API access

YesA documented REST API starts and polls conversations with a bearer key, and the pricing table ticks Cloud APIs on all three plans.OpenHands Cloud provides a REST API that allows you to programmatically interact with OpenHands.docs.openhands.dev/openhands/usage/clou… ↗

MCP server

YesMCP is fully supported on CLI, SDK, local GUI and Cloud, letting the agent consume external tool servers over SSE, SHTTP or stdio.Model Context Protocol (MCP) is an open standard that allows OpenHands to communicate with external tool servers, extending the agent's capabilities…docs.openhands.dev/overview/model-conte… ↗

Bring your own key

YesBring-your-own-key works on every plan, and Enterprise additionally runs on your own infrastructure with your existing provider agreements.Yes! OpenHands Open Source and OpenHands Cloud plans offer support to bring your own LLM keys.openhands.dev/pricing ↗

Buying it for a team

A company can buy it

YesOrganisations buy the Enterprise plan on a custom contract, with multi-user orgs, shared credits and centralised team billing.Centralized team billingopenhands.dev/pricing ↗

Seat model

PartlyNo published per-seat rate at all: the two free plans are single-user, Enterprise is unlimited users and is explicitly sold without seat licensing.Use your own compute resources and LLM API keys. No per-seat licensing.docs.openhands.dev/enterprise.md ↗

Pooled budget

YesOrganisation members draw on one shared credit pool, with an org-wide monthly cap plus per-user budgets admins can override.Create conversations - Start new conversations using the organization's shared credits.docs.openhands.dev/openhands/usage/clou… ↗

Admin controls

YesAdmins constrain the agent's reach: which secrets, tools and domains it may touch, plus enforced default LLMs and spending caps.Define exactly which secrets, tools, and domains each agent can reach. Detect prompt injection. Halt risky actions for human review.openhands.dev/enterprise ↗

Audit log

PartlyFull per-action audit logging is claimed for the Enterprise control plane; the free and Individual tiers carry no such published logging.Complete audit logs for every agent actionopenhands.dev/enterprise ↗

Single sign-on

PartlySAML SSO exists only on self-hosted Enterprise; the comparison table shows it absent from Cloud and from both free tiers.This guide explains how to let users sign in to an OpenHands Enterprise installation with your corporate identity provider (for example Okta,…docs.openhands.dev/enterprise/integrati… ↗

What happens to your code

Opt out of training

PartlyRunning it locally or self-hosted keeps code out of vendor hands entirely, but the cloud privacy policy reserves model training on your content with no opt-out published.we may use content and feedback from your use of the Services to train and tune our AI models to improve the capabilities of our Servicesopenhands.dev/privacy ↗

Data residency

PartlyEnterprise puts code and conversations in your own infrastructure, but no region or jurisdiction is published for the hosted cloud tiers.All code and conversations stay on your infrastructure. Nothing leaves your environment.docs.openhands.dev/enterprise.md ↗

Getting out

PartlyYou can manage and delete stored content, and deletion is by email request; no bulk data export or credit-expiry terms are published.If you require your data to be deleted, please email contact@all-hands.dev.openhands.dev/privacy ↗

Not established: Certifications