Warp Goes Open Source (April 28, 2026): AGPL Client, OpenAI as Founding Sponsor, Oz-Driven Contributions

On April 28, 2026, Warp announced that the Warp client is now open source. The repo lives at github.com/warpdotdev/warp under an AGPL license. OpenAI is the founding sponsor, and Warp is rolling out a contribution model where the heavy lifting is done by Oz agents, with the community handling ideas, direction, and verification rather than submitting traditional pull requests. Warp also expanded support for open-weight models — Kimi, MiniMax, Qwen, plus an auto (open) router that picks an open model per task.

TL;DR

• April 28, 2026: Warp client is now open source at github.com/warpdotdev/warp under AGPL.
• OpenAI is the founding sponsor; GPT models power the Oz-driven contribution workflow.
• Contribution model: agents do the implementation; community provides ideas, direction, and verification — not the traditional PR loop.
• New open-weight model support: Kimi, MiniMax, Qwen, plus an auto (open) router for per-task open-model selection.
• What stays proprietary: the announcement focuses on the client; Oz cloud orchestration is referenced as a separate platform.

Quick Answer

Warp's bet here is twofold. (1) Open-source the client — the part that runs on your machine, the rendering, the terminal core, the agent UX — under AGPL, where derivative network services have to share back. (2) Reframe contribution itself: instead of "PR welcome", contributors describe what should change and an Oz-style agent does the implementation, with maintainers and the community in the verification loop. If you're an agent-tooling builder, the AGPL client is now a reference implementation you can read end to end.

What Was Open-Sourced

"Warp's source code is now available at github.com/warpdotdev/warp with an AGPL license."

The artifact that is open is the Warp client — the GPU-accelerated terminal app you install on your machine. The AGPL choice is meaningful: AGPL extends copyleft to network use, so a hosted service built on the Warp client codebase would have to publish its source under the same terms. That's the same lever MongoDB, Grafana, and others have used to keep cloud providers from re-hosting an open core without giving back.

What is not the same artifact: Oz, Warp's cloud agent platform. The announcement references Oz as the engine driving the new contribution workflow, but Oz cloud orchestration is presented as a separate platform — not the code being open-sourced today. If you previously read our Warp Oz guide, the cloud side described there is unchanged in scope.

OpenAI as the Founding Sponsor

OpenAI is the founding sponsor of the open-source effort. The visible mechanism is that GPT models power the agent workflows Warp is using to land community contributions. There's no exclusivity claim in the announcement — Warp continues to support a multi-model lineup — but the partnership is what funds the agent-first contribution loop rather than a pure volunteer model.

For context on where OpenAI's coding-agent stack stands right now, see our OpenAI Codex April 2026 update.

The Oz-Driven Contribution Workflow

Warp's contribution model is the most unusual part of the announcement. From the post:

"We want agents doing the heavy lifting (coding, planning, testing, etc.) and community members helping with ideas, direction and verification."

Concretely, that means:

• Community contributors open GitHub issues describing intent, scope, and acceptance criteria.
• Oz agents do the planning, implementation, and testing.
• Warp maintainers and community reviewers verify, redirect, and accept work — collective knowledge management rather than a traditional PR queue.

If this works as described, Warp will be one of the first mainstream open-source projects to make agent-implemented, human-verified the default path rather than a side experiment. It also pairs naturally with the AGPL choice: the more the implementation is agent-produced and shared, the higher the leverage of keeping it under copyleft.

Wider Open-Weight Model Support

In the same release, Warp expanded its model lineup with three open-weight options and a new auto-router:

The auto (open) mode mirrors the per-task routing pattern used by Cursor's auto mode and Claude Code's auto mode for Max users, but constrained to open-weight models. For developers in regulated environments where data residency or model provenance matters, that's a meaningful lever.

Why This Matters for AI Coding Tools

A few signals worth pulling out:

• The AGPL terminal client is now a reference implementation. Anyone building an agentic terminal — or evaluating one for a regulated stack — has a real codebase to read instead of inferring behavior from screenshots.
• Agent-first contribution lowers the barrier from "I can write Rust + know your codebase" to "I can describe what I want and verify what came back." That widens the contributor pool, but only if the verification loop is rigorous enough to keep quality high.
• The OpenAI sponsorship is a notable cross-vendor move: OpenAI underwriting an open-source agentic terminal that supports many other model vendors, including direct competitors.
• AGPL is the right hammer for the cloud-fork problem, but it also raises the bar for enterprises that want to embed Warp's client code in their own paid tools — something to check with your license team before forking.

How This Compares

Sources

• Warp blog — Warp Is Now Open Source (April 28, 2026): warp.dev/blog/warp-is-now-open-source
• Warp client source: github.com/warpdotdev/warp

For broader context on agentic terminal tools see our Warp Oz guide, AI CLI coding tools roundup, and best open-source AI coding tools.