Warp open-sourced its terminal client on Tuesday, dropping the Rust codebase on GitHub under an AGPL v3 license with OpenAI signing on as founding sponsor. The repo crossed 30,000 stars within hours of the announcement and now sits north of 37,000. What it doesn't open is the part of the business that makes money.
Agents code, humans verify
The interesting bit isn't the license. It's the contribution model. Warp wants community members to file issues, spec features, and review pull requests, but the actual code writing is meant to happen via Oz, Warp's cloud agent orchestration platform, which is staying proprietary. Maintainers tag issues "ready-to-spec" or "ready-to-implement" and agents pick them up. The GitHub repo documents the flow in CONTRIBUTING.md.
"We're excited to support experiments that explore how AI can help maintainers and contributors collaborate more effectively at scale," said Thibault Sottiaux, Engineering Lead at OpenAI, in the company's announcement post. Which is the kind of thing you'd expect a sponsor to say. The harder question, and one I don't have a clean answer for, is whether contributors who can already write Rust will want to file tickets for a GPT-5.5 agent to implement when they could just write the code themselves.
Founder Zach Lloyd's framing is that the bottleneck has shifted. Code-writing is no longer the hard part; deciding what to build and verifying it works is. That's a defensible position if you accept the premise. Plenty of contributors won't. And does anyone really enjoy writing spec docs for an AI?
What's actually open
Under AGPLv3: the client codebase, in Rust. Under MIT: the UI framework crates (warpui and warpui_core), which is more permissive and lets people reuse those pieces without copyleft obligations. Themes and the workflows repo were already open before this.
What's not open: Oz itself, the server-side infrastructure, and the commercial features that drive Warp's revenue. The $12/month Pro tier still exists. Enterprise features still cost money. So this is a client-only release with the orchestration layer that does the actual agent work kept behind a paywall, which is fine, that's the business model. It just complicates the "community-driven" framing.
AGPLv3 was almost certainly chosen to prevent cloud providers from running a hosted Warp clone without contributing changes back. Same defensive pattern MongoDB and Elastic used. The GPL purist crowd will appreciate the copyleft. The "actually open" crowd will note that everything interesting is still proprietary.
The Alacritty question
Hacker News debate turned quickly to a sore spot. Warp was built on top of Alacritty, the open-source Rust terminal emulator, and went on to raise around $50 million in venture funding without contributing meaningfully back. Several commenters made this point, with varying degrees of politeness. One suggested the company could donate some of that $50 million if it was actually grateful for the foundation it built on.
An AGPL release doesn't address that complaint. It governs what happens going forward; it does not retroactively support the projects Warp leaned on to get here. There's a competitive angle too. Ghostty, written in Zig and fully free, has been picking up serious momentum as a lighter-weight alternative. Open-sourcing the client now reads, depending on who you ask, as a genuine contribution or as a defensive move once terminal emulators became commodity software.
Watch the dashboard
Warp set up build.warp.dev as a live dashboard showing issue triage, agent activity, and what's available to pick up. That's the real test. If the community drives direction and outside contributors actually get features shipped without learning the codebase, the model holds up. If agents just implement Warp's existing roadmap with the community supplying flavor text, it's a different story.
One more thing to watch: Warp says the Oz workflow is now available to other open-source projects, not just its own. Whether any sizable project actually adopts it, or whether this stays a Warp-only experiment with OpenAI's backing, will say something about how far "open agentic development" travels as a model.
