Warp Open-Sources Its Agentic Terminal Under AGPLv3 With OpenAI as Founding Sponsor and an Agent-First Contribution Model

Overview

Warp, the Rust-based AI terminal that has been promising to go open source for years, finally released its client codebase on GitHub on April 28, 2026, according to the company’s announcement. The release places the bulk of the client under AGPLv3 while the UI framework crates ship under MIT, and it positions OpenAI as the repository’s founding sponsor in a contribution workflow that explicitly asks AI agents — not humans — to do most of the coding.

The announcement closes a chapter that, in the words of FOSS Force, saw the company “flirting — and not flirting — with making Warp open source for about four years.” It also drops Warp into a market where rival AI coding tools, including the Cursor and JetBrains environments The Machine Herald has previously covered, have largely stayed proprietary.

What Was Released

The code now lives at github.com/warpdotdev/warp, where the repository describes itself as “an agentic development environment, born out of the terminal.” The bulk of the codebase is licensed under AGPL-3.0, while the UI framework crates warpui_core and warpui ship under the MIT license, as detailed by It’s FOSS. The project is written almost entirely in Rust and runs on macOS and Linux, per the repository metadata. It’s FOSS adds that Windows is also supported.

Uptake has been brisk. byteIOTA reported that the repository “climbed to #2 on GitHub trending” in the week following the announcement and accumulated more than 37,000 stars within days; the same outlet pegs Warp’s existing user base at more than 700,000 developers. The GitHub project page now shows the star count above 57,000.

Alongside the source drop, Warp expanded its supported model lineup. The official post lists Kimi, MiniMax and Qwen as newly available providers and introduces an “auto (open)” routing option that, the company says, “picks the best open model for a task,” a feature also noted by Help Net Security.

What Did Not Open

The code release stops well short of a fully self-hostable Warp. FOSS Force notes that the proprietary side of the stack — Warp Drive and the Oz orchestration platform that powers the company’s agent workflows — remains closed and commercial. Help Net Security similarly reports that contributions are routed through Oz, a cloud orchestration system, with agent workflows running on OpenAI’s GPT models.

Kilo Blog observes that the practical effect is a split between code and platform: “AI features depend on Warp’s backend services, preventing fully self-hosted experiences,” and bring-your-own-key access for Anthropic, Google and OpenAI keys is available only on paid plans. Within days a community fork, OpenWarp, emerged that strips those constraints and lets users wire in any OpenAI-compatible endpoint, byteIOTA reported.

An Agent-First Contribution Model

What makes the release unusual is not the license but the contribution workflow. Warp is asking AI agents, not humans, to write most of the new code. In the company’s framing, the new model centers on “agents doing the heavy lifting (coding, planning, testing, etc.) and community members helping with ideas, direction and verification,” according to Warp. It’s FOSS describes the same arrangement: humans focus on “ideas, spec work, and review” while AI agents handle implementation, it reported.

Founder and chief executive Zach Lloyd was blunt about the rationale. Speaking to It’s FOSS, he said: “Open-sourcing is fundamentally coming from our desire to build a successful business. We are competing with other highly funded, closed-source competitors, and we think opening and providing the resources for the community to improve Warp is a smart way for us to accelerate product development,” in remarks quoted by the outlet. In a Hacker News thread the same day he added: “I actually think we can build a better product, more quickly if we build it with our community + agents,” per the discussion.

FOSS Force quoted Lloyd on the underlying premise: “The biggest bottleneck to development is no longer writing code — it’s all the human-in-the-loop activities around the code.”

OpenAI’s role is more than symbolic. The company is the repository’s founding sponsor and supplies the GPT models that run Oz, Warp confirmed. Thibault Sottiaux, an engineering lead at OpenAI, said in the announcement: “Open source has long been central to how developers learn, build, and push the field forward. We’re excited to support experiments that explore how AI can help maintainers and contributors collaborate more effectively at scale.”

Reception and Loose Ends

The reception on Hacker News was mixed. One commenter welcomed the license choice: “I was very surprised to see they chose my favorite free software license — the AGPLv3.” Others were less complimentary about Warp’s roots. A user named shimman criticized the company for “tweaking an Alacritty fork then using that to get a $50million venture round then giving zero money,” though Alacritty contributor nixpulvis responded that he had “absolutely no hard feelings,” the thread shows. Telemetry concerns also resurfaced: one user reported that blocking the terminal’s internet access caused it to refuse to start, the discussion records.

Warp’s decision arrives during a period of churn in open source stewardship. The Machine Herald reported in late April that Intel had archived its open source evangelism hub and several community projects, and the broader industry has been wrestling with the question of who pays for the work AI agents are increasingly being asked to do. Warp’s answer — sponsored agents writing code in exchange for human review cycles — is one model. Whether contributors accept the trade-off, and whether the proprietary backend keeps the lights on as Lloyd predicts, will become clearer as the repository matures.

What We Don’t Know

• The detailed economics of OpenAI’s sponsorship were not disclosed in any of the cited announcements.
• The split between code and platform leaves open how much functionality a fully self-hosted build of the open-source client would actually retain.
• Whether the agent-first contribution workflow produces accepted pull requests at scale, and how human reviewers will be compensated for verification work, has yet to play out in practice.