OpenHands
Also known as: OpenDevin
MIT licensed open source agent platform whose Agent Canvas self hosts and automates coding agents across local, VM, cloud and enterprise backends, running its own agent or third party harnesses like Claude Code, Codex and Gemini CLI over ACP.
OpenHands is an open source platform for autonomous AI software development, released under the MIT license. It began as OpenDevin, a community answer to closed autonomous coding agents, and has grown into an open agent you can run yourself. Its premise is that agentic development is too important to be controlled by a few companies, so the whole stack is built in the open.
Rather than suggesting code as you type, OpenHands runs agents that do the work a developer does. They plan a task, write and edit files across a codebase, run terminal commands, install dependencies, browse the web, run tests and open pull requests, iterating until the task is complete or they reach a defined stopping point. You can hand it a plain language task or point it at a GitHub issue and let it locate the code and submit a fix. Every agent runs inside an isolated sandbox, typically a Docker container, so it can act freely without risking the host machine.
OpenHands is model agnostic. It connects to any major model, including Claude, GPT and Gemini, or to local models, so teams choose their own provider and avoid lock in. Its control center, Agent Canvas, turns coding agents into an always on, self hosted engineering team. It runs the built in agent or third party ones such as Claude Code, Codex, Gemini CLI, Pi and OpenCode through the Agent Client Protocol, across local, remote, Docker, Kubernetes and cloud backends. Automations run on a schedule or fire from GitHub events and webhooks, prebuilt automations take an issue to a pull request or review one, and a Critic can score the agent's actions and send it back to refine its work.
For builders, an open agent SDK exposes the same engine that powers Agent Canvas and the hosted cloud, letting developers add custom tools and strategies or embed agents in their own products. For organizations, Teams is a self hosted plan with multi user identity, SSO and SAML, role based access and budgeting, and Enterprise installs on a VM or into the customer's own Kubernetes cluster inside its VPC on AWS, Azure or Google Cloud, with Model Context Protocol support throughout. Because the platform is MIT licensed, teams can self host, fork and modify the whole stack and keep the workflow on their own infrastructure.
From our cloud coding agents video
Devin, Factory, Jules, OpenHands: What Each Cloud Coding Agent Does Best
Our video on the 11 cloud coding agents in the index names what each one does best. For OpenHands, that is the self hosted fleet.
The player starts at OpenHands's part, 5:02 in. Recorded in October 2026; the profile here is live and updates as research is verified.
Vendor details
Canonical URL
https://www.openhands.dev
Category
Coding agent
Funding status
Independent, built by the OpenHands team (All Hands AI). MIT licensed open source, formerly OpenDevin. The site shows about 92,000 GitHub stars. Commercial model is a self hosted Teams plan plus an Enterprise tier with the Agent Control Plane. Ecosystem partnerships with NVIDIA for Nemotron models and AMD for Instinct GPU deployment; a member of the Open Secure AI Alliance.
Company status
independent
Use cases & customers
Target customers
Deployment options
Integrations
MCP client support ships on every plan, with the agent connecting out to external MCP servers and using their tools automatically. Built in integrations cover GitHub, GitLab, Bitbucket and Azure DevOps for source control, Jira Cloud and Jira Data Center for issues, and Slack for messaging, and GitHub events and custom webhooks trigger automations. Agent Canvas connects to multiple agent backends and runs third party harnesses including Claude Code, Codex, Gemini CLI, Pi and OpenCode through the Agent Client Protocol. A Python SDK supports custom tools, multi agent delegation and remote workspaces on Docker, VMs or Kubernetes, with Modal and DigitalOcean Managed Agents as remote backends and ngrok or Tailscale for remote access.
In practice
You want an autonomous coding agent on your own hardware with your own model. OpenHands is open source and model-agnostic, running coding agents in a sandboxed Docker container against whatever LLM you choose.
You'd rather hand off a GitHub issue than write the fix yourself. OpenHands takes the issue, locates the relevant code, makes the changes, runs the tests, and opens a pull request for you to review.
Your code can't go to a third-party service. OpenHands self-hosts on your own infrastructure, runs agents in isolated sandboxes, and offers a private-cloud enterprise deployment with access controls and budgeting.
Sources & related URLs
Related / legacy domains
Agentic Index coverage score
13.5 / 14 capabilities · 96%
| Integrations & Tool Calling | Full |
|---|---|
|
MCP support ships on every plan as a client, with the agent connecting out to external MCP servers and using their tools automatically. Built in integrations cover GitHub, GitLab, Bitbucket (Cloud and Data Center) and Azure DevOps for source control, Jira Cloud and Jira Data Center for issues, and Slack for messaging, and the pricing page lists built in OAuth integrations on Teams and Enterprise with select MCP only on the free plan. The agent runs terminal commands, installs dependencies and browses the web inside its sandbox. Sourcedocs.openhands.dev/enterprise/integrations/overview, overview/model-context-protocol and openhands.dev/pricingread 2026-10-28 |
|
| Workflow Orchestration | Full |
|
Agent Canvas runs parallel conversations in one workspace and orchestrates multiple agent backends, running the built in OpenHands agent or Claude Code, Codex, Gemini CLI, Pi and OpenCode through the Agent Client Protocol. The agent plans a task, then writes and edits files, runs terminal commands, browses the web, runs tests and opens pull requests until done. The SDK's Task Tool Set delegates work to specialized sub agents, file based agents define them as Markdown, and the Software Factory use case chains four automations to take an issue through development, review, testing and merge. Sourcedocs.openhands.dev agent-canvas/acp-agents, sdk/guides/task-tool-set and use-cases/software-factoryread 2026-10-28 |
|
| Knowledge Grounding & RAG | Partial |
|
The agent reads and navigates the codebase directly inside its sandbox, locating relevant code from a plain language task or a GitHub issue, browses the web for external information and reaches connected data sources through MCP servers. Repository skills in a .openhands directory and path triggered rules give it standing guidance, and Persistent Memory carries notes forward. The documentation index lists no codebase index, embeddings or retrieval layer. Sourcedocs.openhands.dev customization/repository, overview/skills and llms.txtread 2026-10-28 |
|
| Human Oversight & Guardrails | Full |
|
A confirmation policy decides when an action needs the user's approval before it runs: AlwaysConfirm gates every action, ConfirmRisky gates only actions a security analyzer rates risky, and NeverConfirm runs unattended. A pending action leaves the agent waiting, the user can reject it with feedback so the agent tries another approach, and the agent server API accepts or rejects pending actions. Hooks block dangerous commands before they run, every agent runs in an isolated sandbox, and work lands as a pull request for normal review. Scheduled automations run unattended by design. Sourcedocs.openhands.dev/sdk/guides/security, sdk/arch/security and customization/hooksread 2026-10-28 |
|
| Security, Identity & Governance | Full |
|
Teams and Enterprise add SAML and SSO, multi user identity and role based access control, budgeting controls and centralized audit logs, and Enterprise documents SAML SSO setup, organization roles and permissions and self hosting in the customer's own VPC on AWS, Azure or Google Cloud. The SDK adds a security analyzer and configurable security policies, and every agent runs in an isolated sandbox. The Trust Center linked from the site sits on Vanta and refuses automated reads, and no readable page names a SOC 2 or ISO attestation, so the compliance half rests on the documented controls with that gap stated. Sourceopenhands.dev/enterprise, openhands.dev/pricing, docs.openhands.dev/enterprise/integrations/saml-sso and organizations/roles-permissionsread 2026-10-28 |
|
| Observability & Auditability | Full |
|
Built in OpenTelemetry tracing records agent execution steps and each tool call, exportable to any OTLP compatible backend the customer runs (Laminar, MLflow, Honeycomb, Jaeger, Datadog, New Relic) with transcript, tree and timeline views of prompts, tool calls, outputs and nested agent activity, and a metrics guide tracks tokens, cost and latency. OpenHands Enterprise sends conversation traces to the customer's own OTLP platform, deploys Laminar for trace analysis, and lists centralized audit logs and per user usage and cost attribution. Sourcedocs.openhands.dev/sdk/guides/observability, enterprise/integrations/observability-platforms, enterprise/analytics and openhands.dev/enterpriseread 2026-10-28 |
|
| Memory & State Persistence | Full |
|
Persistent Memory is an opt in two tier store the agent maintains itself: a user tier under ~/.openhands/memory for knowledge across projects and a project tier under the workspace's .openhands/memory, each holding a curated MEMORY.md index injected at session start and daily logs read on demand. The agent is instructed to record durable facts, merge duplicates and prune stale entries, the combined index is capped at about 6,000 characters with the oldest lines dropped first, and the files are plain Markdown that people can review, edit or delete at any time. AGENTS.md holds instructions, separate from memory. Sourcedocs.openhands.dev/sdk/guides/persistent-memoryread 2026-10-28 |
|
| Deployment & Data Residency | Full |
|
Agent Canvas connects to local, remote, Docker, Kubernetes (Helm), Modal, DigitalOcean Managed Agents and OpenHands Cloud backends, and every plan on the pricing page is self hosted. OpenHands Enterprise installs on a Linux VM with the Replicated installer, on AWS, Google Cloud or Azure VMs, into the customer's own EKS, AKS or GKE cluster with Helm, or with Docker Compose, keeping code, data and execution inside the customer's VPC. The platform is MIT licensed, so the whole stack can be forked and modified. Sourcedocs.openhands.dev agent-canvas/backends, enterprise/vm-install/generic-vm, enterprise/k8s-install/index and openhands.dev/pricingread 2026-10-28 |
|
| Prebuilt Agents, Templates & Packs | Full |
|
Prebuilt automations ship as named setups the customer installs: Issue to Pull Request, GitHub PR Review Assistant, GitHub Repository Monitor and Slack Channel Monitor. Plugins bundle skills, hooks, MCP servers, agents and commands into packages in the portable Agent Plugins format, Agent Canvas browses and installs them, Enterprise can enable a Plugin Marketplace of community plugins, and an official global skill registry lives at github.com/OpenHands/extensions. Sourcedocs.openhands.dev agent-canvas/prebuilt-automations, overview/plugins, enterprise/plugin-marketplace and overview/skills/publicread 2026-10-28 |
|
| Triggers & Channel Coverage | Full |
|
Automations run on a cron schedule or fire from GitHub events and custom webhooks, and prebuilt monitors watch a GitHub repository or a Slack channel and start agent actions on events. The enterprise page describes the path as an issue opened, an agent fired, an isolated container and a pull request opened with no human in the loop, and Jira, Azure DevOps and GitLab webhooks are configured in Enterprise. Sourcedocs.openhands.dev automations/overview, automations/event-automations, agent-canvas/prebuilt-automations and openhands.dev/enterpriseread 2026-10-28 |
|
| Model Flexibility & Routing | Full |
|
OpenHands connects to any model LiteLLM supports, with provider guides for OpenAI, Anthropic through Bedrock, Azure, Google Gemini and Vertex, Groq, Moonshot, OpenRouter, DigitalOcean and local servers such as Ollama, vLLM, LM Studio and SGLang, plus its own OpenHands LLM provider. LLM Profiles let a developer switch models mid conversation, the SDK routes requests across models by task and falls back to alternates on failure, and Enterprise chains to an external LiteLLM or Bifrost gateway. Sourcedocs.openhands.dev usage/llms/llms, agent-canvas/llm-profiles, sdk/guides/llm-task-routing and enterprise/integrations/external-llm-gatewaysread 2026-10-28 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
A Python Software Agent SDK exposes the same engine that powers Agent Canvas and the hosted cloud, with custom tools, MCP, sub agent delegation, hooks and configurable security. The OpenHands Cloud REST API and the Agent Server REST API, both with published OpenAPI specs, start and drive conversations, sandboxes, events, secrets and settings with API keys, the Agent Server can be called through an OpenAI compatible endpoint, and the whole platform is MIT licensed and forkable. Sourcedocs.openhands.dev/sdk, usage/cloud/cloud-api and sdk/guides/agent-server/overviewread 2026-10-28 |
|
| Testing, Debugging & Optimization | Full |
|
A Critic scores the agent's actions in real time with an LLM critic model and drives iterative refinement until a quality threshold is met, in Agent Canvas and the SDK, and a Goal Completion Loop drives a conversation toward a verifiable objective with a judge. A benchmarks repository provides standardized evaluation pipelines for OpenHands agents (SWE-bench inference and scoring commands, parallel runs in isolated containers), a skills monitoring guide covers evaluation metrics and dashboards for skills in production, and OpenHands publishes the OpenHands Index, its own scoring of model and SDK combinations across categories of software engineering work. Sourcedocs.openhands.dev agent-canvas/critic, sdk/guides/convo-goal, overview/skills/monitoring and index.openhands.devread 2026-10-28 |
|
| Browser & Computer Use | Full |
|
The agent browses and interacts with web pages through a Browser Use tool inside its sandbox, browser sessions can be recorded and replayed with rrweb, the agent server exposes a noVNC desktop URL, and the Automated QA use case validates pull request changes by running the software rather than only reading code or running tests. Sourcedocs.openhands.dev sdk/guides/agent-browser-use, sdk/guides/browser-session-recording and use-cases/qa-changesread 2026-10-28 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
Multi repository conversations arrive in OpenHands Enterprise 0.103.0, so a user can start one conversation across several repositories from the web app, Slack, Jira or Jira Data Center. A Jira ticket that names several repositories now starts a single conversation, and the agent can clone private repositories from any connected git provider. Administrators can also restrict outbound traffic from agent sandboxes with an opt in Kubernetes network policy.
Bears on: Agent capability
View sourceA new cortex-training-rl skill lets the OpenHands agent set up, launch and monitor reinforcement learning runs that tune an open weight model inside a team's own agent harness, with training on Snowflake Cortex Training GPUs. Each rollout runs in its own Snowflake Cortex Sandbox, and the agent reports reward metrics, plots the reward curve and releases the GPUs when the run ends.
Bears on: Integrations
View sourceCustom webhook sources can now trigger automations in Agent Canvas 1.35.0, through a new Events page under Automate where teams list and register them. An automation can also keep its sandbox alive after a run finishes.
Bears on: Workflow orchestration
View sourcePricing
Free as open source (Agent Canvas, MIT). Teams is $500 a month for up to 10 users, plus $50 a month for each extra seat, with a 30 day trial. Enterprise is custom priced. Every plan runs on the customer's own model key.
A flat monthly team plan for up to 10 users with a charge for each extra seat, or a custom Enterprise contract. Model inference is paid separately on the customer's own key.
Included quota
Agent Canvas is free and open source under MIT, self hosted on a laptop or shared VM for a single user or shared identity on the customer's own model key, with select MCP integrations only. Teams covers up to 10 users and up to 5,000 sessions a month, and adds multi user identity, SSO and SAML, RBAC, built in OAuth integrations and budgeting controls. Enterprise is custom, with unlimited seats and organizations and uncapped usage.
What is public
OpenHands (openhands.dev, formerly OpenDevin) publishes three self hosted plans. Agent Canvas is free and open source under MIT, for a single user or shared identity on the customer's own model key. Teams is $500 a month for up to 10 users, plus $50 a month for each extra seat, and is capped at 5,000 sessions a month. It adds multi user identity, SSO and SAML, RBAC, built in OAuth integrations and budgeting controls, and comes with a 30 day trial. Enterprise is custom priced, with unlimited seats, unlimited organizations, uncapped usage, an enterprise agreement, SLA based support and a customer success manager.
Billing mechanics
OpenHands charges a flat monthly price for Teams and a custom contract for Enterprise, and adds nothing on model usage. Every plan runs on the customer's own model key, so inference is billed by the customer's provider.
Cost watchouts
Model inference is not included on any plan and scales with task volume and model choice. Teams is capped at 5,000 sessions a month. Self hosting takes setup work (Docker or Kubernetes, security hardening, model configuration). Enterprise rates are quoted by sales only.
Variable cost rationale
The plan price is flat: $500 a month for Teams up to 10 users, plus $50 for each extra seat. Model inference is the variable part, paid on the customer's own key, so spend grows with task volume and the model chosen. The Teams cap of 5,000 sessions a month limits heavy use rather than billing for it. Enterprise is a custom contract with no published rate.
Additional watchouts
Model costs sit outside every plan, and cheap models give weaker agent results. Teams is capped at 5,000 sessions a month. Self hosting takes setup work. Tasks need tight scoping and human review, because ambiguous specs yield partial or incorrect solutions.
Overage / add-ons
Teams adds $50 a month for each seat past 10 and is capped at 5,000 sessions a month rather than billed for overage. Enterprise usage is uncapped under a custom contract. Model usage is billed by the customer's own provider.
Sales call required
Mixed (some tiers require a call)
Free / trial
Free open source Agent Canvas for a single user, and a 30 day trial of Teams.
Lowest paid plan
Teams at $500 a month for up to 10 users
Commercial notes
OpenHands began as OpenDevin, a community project started in early 2024 and co-created by CMU's Graham Neubig, and was rebranded under All Hands AI. It raised an $18.8M Series A led by Madrona in November 2025. The site shows about 92,000 GitHub stars, and the product is pitched as the self hostable autonomous coding agent a team runs itself, with other agents (Claude Code, Codex, Gemini CLI) runnable inside it.
Key ambiguities
The pricing page sells three self hosted plans (Agent Canvas, Teams and Enterprise) and no hosted plan, while the site's navigation still links OpenHands Cloud and an Access Cloud button opens app.all-hands.dev. Whether a hosted plan for individuals is still sold is not stated on the pricing page.
Cancellation / refund
The open source edition is free with no contract. Teams is a monthly plan with a 30 day trial. Enterprise is a custom contract.
Support SLA / resale
Open source users get community support. Enterprise adds priority support, a shared Slack channel and a named customer engineer.
Missing data
Enterprise pricing is not published. The pricing page does not say whether the hosted OpenHands Cloud individual plan is still sold.
Related vendors
- Cognition — Maker of Devin, an autonomous AI software engineer
- 10Web — Agentic website platform whose AI agents build, host and manage…
- AgentUI — Managed AI app builder for operations teams that generates and hosts…
- Aider — Open source, model agnostic terminal coding agent that edits your…
- Amp — Amp is a coding agent from Amp Frontier Corporation that runs each…
- Anthropic Claude Code — Anthropic's agentic coding system across terminal, desktop, IDE, web…
Alternatives to OpenHands
The closest documented capability profiles to OpenHands among coding agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Replit Agent13.5 / 14Matches OpenHands across all 14 documented capabilities
- Base4413.0 / 14A lighter documented profile than OpenHands
- OpenAI Codex13.0 / 14A lighter documented profile than OpenHands
- Anthropic Claude Code12.5 / 14A lighter documented profile than OpenHands
- Capy12.5 / 14A lighter documented profile than OpenHands
- Cognition13.5 / 14Fuller documented coverage on Knowledge Grounding & RAGOpenHands vs Cognition →
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded