Comparison
Shipwright vs OpenHands
OpenHands is the open-source category leader in autonomous coding agents — MIT-licensed, model-agnostic, and it already runs a real delivery pipeline: scheduled automations, PR review, and an Agent Canvas dashboard to manage it all. This is not an open-vs-commercial story the way Devin or Factory are — both Shipwright and OpenHands ship MIT cores. The real difference is what's enforced by default versus what's optional, and how much of the richer functionality lives behind their commercial offering rather than in the open-source core. Here is the honest, sourced comparison.
Facts verified as of September 5, 2026, from OpenHands' own primary sources (linked below). Re-checked immediately before this page ships: vendor terms and product surfaces change.
Dimension by dimension
| Dimension | Shipwright | OpenHands |
|---|---|---|
| License | MIT, whole harness forkable. | MIT. [source] |
| Deployment | Self-host: Docker or your own Kubernetes. | Local or a self-hosted VM backend by default. A managed hosting option and a separate self-hosted commercial option are also available. [source] |
| Models | Claude Code, tuned end to end. | Agnostic (100+ providers), plus ACP support that lets it launch and drive Claude Code, Codex, or Gemini CLI directly as a subprocess agent. [agnostic] [ACP] |
| Task queue | Shared, assignable task-store queue, cron-native. | Automations: schedule daily / weekly / custom-interval tasks. Each run spawns a fresh sandbox with the configured LLM, secrets, and integrations already wired in. [source] |
| Team visibility | Slack plus the PR trail. | Agent Canvas: a dashboard for creating and managing automations. Deeper team-wide visibility — shared workflows, audit logs, cost guardrails — sits in their commercial offering, not the open-source canvas. [source] |
| Policy controls | Per-repo, per-skill. | None in the open-source core. Role-based access control (RBAC) and Keycloak-based multi-user management are part of their commercial offering only. [source] |
| Tests-first | Enforced: tests land in the same PR, written first, CI-green to merge. | Opt-in QA agent, invoked manually via a slash command or an optional CI automation you wire up yourself. It explicitly does not run the test suite, linters, formatters, or type-checkers — a supplementary check, never a merge-blocking gate. [source] |
| Slack workflow | First-class: the whole operating loop runs in Slack. | Yes — Slack, GitHub, and GitLab integration is available out of the box; a one-click setup flow is part of their commercial offering. [source] |
| Open source | Yes, fork and audit every line. | Yes, MIT-licensed core, with a separate commercial offering layered on top for hosting and team management. [source] |
The architectural contrast
The resemblance here is closer than with any other competitor: both projects are MIT, both are built to open real PRs against your repository rather than hide the work in a vendor UI, and both already run a scheduled, queue-driven delivery pipeline. Where they diverge is what each project treats as a default versus an opt-in extra, and where the line falls between the open-source core and each project's own commercial offering.
Where OpenHands is ahead, plainly. Model and agent flexibility is OpenHands' clearest strength: it supports 100+ providers, and its ACP (Agent Client Protocol) integration goes further than most — instead of calling an LLM directly, OpenHands can spawn Claude Code, Codex, or Gemini CLI as an external subprocess agent and relay each turn to it, so you can literally drive Claude Code from inside OpenHands. Its VCS reach is also broader today: OpenHands connects to GitHub, GitLab, and Bitbucket out of the box, with broader reach including Azure Repos. It also offers a managed hosting option for teams that don't want to run any infrastructure themselves, and OpenHands' team publishes their own agentic-coding benchmark results; Shipwright does not. If model breadth, VCS coverage, managed hosting, or a published benchmark matter to your decision, OpenHands has them today and Shipwright does not. We are not going to pretend otherwise.
The centerpiece difference: tests-first is enforced in Shipwright, optional in OpenHands. OpenHands ships a QA agent, but it is invoked manually or through a CI automation you configure yourself — and by OpenHands' own documentation, "the QA agent does not run the test suite" (that's CI's job), nor does it run linters, formatters, or type-checkers. It is a supplementary, discretionary layer, never an enforced pre-merge gate. Source: docs.openhands.dev QA-changes use case. In Shipwright, tests are not a layer you opt into: every task ships its tests, written before the implementation, and CI must be green before merge — on every task, with no configuration required.
Where Shipwright is ahead. Two things OpenHands does not enforce. First, tests are not a discretionary agent you invoke — they are the default path for every change, gated by CI. Second, plan approval is the standard operating mode for all work: a human reviews and approves the plan before the agent builds, not after it has already opened a PR. And underneath both, Shipwright's control plane, task queue, and review flow all run on infrastructure you own — nothing about the pipeline itself depends on a commercial offering layered on top.
So the choice is not "which one is open source." Both are. It is whether you want the broadest model and VCS flexibility available today with a mature ecosystem behind it, or a harness that makes tests-first and plan-approval the default on every task and keeps the entire pipeline — not just the agent — on infrastructure you control.
Choose OpenHands when
Choose OpenHands when
- You need maximum model and agent flexibility — 100+ providers, or ACP support to drive Claude Code, Codex, or Gemini CLI directly.
- Your team spans GitHub, GitLab, and Bitbucket today, and you want that VCS breadth (with wider reach including Azure Repos) out of the box.
- You want a managed hosting option and don't want to run any infrastructure yourself.
- You want the most mature open-source agent ecosystem available right now, with published benchmark results to evaluate against.
Choose Shipwright when
- You want tests written first and enforced as a merge gate on every task, not a discretionary QA agent you invoke manually.
- You want plan approval as the default path on every task, not an extra step you have to configure.
- You have committed to Claude Code and want a pipeline tuned for it, not averaged across 100+ providers.
- You want the whole harness — control plane included — running on infrastructure you own, MIT throughout, with nothing behind a separate commercial offering.
Economics
Shipwright: free, MIT, no seats or paid plans. OpenHands' open-source core is likewise free under MIT; broader hosting and team-management capabilities are available through their commercial offering — see the vendor's site for current terms, which change over time. Either way, you bring your own model credentials and pay for the compute and model usage you actually consume.
Try it
Install the plugin into Claude Code and run a task end to end. Free, MIT, runs on your own infra.
/plugin install shipwright@app-vitals/shipwright