On a spring morning in late April, Warp Terminal did something unusual for a developer tool approaching a million users. It didn't just open-source its code—though it did that, releasing the terminal client under an AGPLv3 license and its internal Rust UI framework under MIT. Instead, the company effectively asked its new community contributors to step aside and let the machines go first.
The twist, as CEO Zach Lloyd explained it, comes down to workflow. Warp is funneling contributions through Oz, its cloud-based agent orchestration platform, in what Lloyd termed an "agent-first" approach. GPT models now handle the grunt work: triaging issues, drafting specifications, implementing changes, even conducting preliminary code reviews. Human developers still make the final call, but only after the AI has had its turn.
Within 48 hours of the April 28 announcement, the GitHub repository had pulled in 45,300 stars, 2,800 forks, and north of 3,000 issues. A public dashboard showed improvements underway by April 30, with a live feed of pull requests and Oz sessions scrolling past in real time. The response was immediate, fervent—and, in places, skeptical.
Why Now, Why This Way
Lloyd's explanation for the timing sidestepped the usual idealism about community and collaboration. "Open-sourcing is fundamentally coming from our desire to build a successful business," he wrote. The company faces well-funded competitors with closed codebases, and opening up, Lloyd argued, represents "a smart way for us to accelerate product development."
That business, he clarified in a Hacker News thread shortly after, isn't actually the terminal. "Our business is our agent and orchestrator, not our terminal," Lloyd noted—a distinction that explains which parts Warp is willing to give away and which it plans to monetize. The terminal becomes the storefront; Oz, the proprietary engine humming behind it.
The open-source release covers the Warp client codebase. Cloud-based enterprise features remain locked down for now, and reporting from Fast Company suggested certain desktop components tied to specific clients or model testing may also stay closed. It's selective transparency, calibrated around competitive positioning.
The Contribution Gauntlet
Here's where Warp's model diverges. Feature requests don't go straight to code. They move through stages, with AI agents tackling the early legwork.
An issue gets tagged "ready-to-spec" before anyone drafts product or technical specifications. Those specs land in a tidy directory structure—specs/GH<issue-number>, complete with product.md and tech.md files—written by agents, then reviewed by human subject matter experts. Only after approval and a "ready-to-implement" label can pull requests proceed.
The agents run on OpenAI's GPT models. They triage, they spec, they implement, they review. Warp's human maintainers step in for what the company calls "SME review"—the final judgment call before merging. It's a kind of assembly line, perhaps, though one where the workers are probabilistic and the foreman still human.
Whether this accelerates development or just introduces a new bottleneck depends on whom you ask. Warp clearly believes the former. The community hasn't quite made up its mind.
What's Released, What Isn't
Alongside the code, Warp expanded model support on April 28. The terminal now offers Kimi, MiniMax, and Qwen, plus an "auto (open)" feature that routes tasks to the best-suited open-source model.
Configurability ranges from minimal—"just a terminal," as Warp puts it—to what the company calls an ADE, or agentic development environment, complete with diff view, file tree, and embedded agents. A new settings file lets users control preferences programmatically.
Oz itself, however, remains firmly in the cloud. While Warp has "open-sourced a number of skills," according to a February launch post, the orchestration platform is positioned as SaaS, not community code. A Wikipedia entry claiming Oz went open-source in April appears to conflict with primary sources and should be read with caution.
Friction in the Forums
Hacker News threads lit up with the predictable tensions. Telemetry. Bloat. Network requests at startup. One Warp engineer acknowledged that builds from the open-source repo strip telemetry and crash reporting entirely—but that didn't stop the questions.
Then came the bug. A "very bad bug," in the words of one engineer, that re-enabled telemetry after restart for some new users. A fix shipped quickly, but the incident underscored the scrutiny that trails any open-source debut.
An April 28 AMA on Reddit fielded questions about contribution workflows, local model support, and the now-lifted login requirement (Warp dropped that barrier in November 2024). The team confirmed that local-model support could be queued for community work. Community members raised requests for custom OpenAI-compatible endpoints in the days following the announcement.
Requests for tmux integration, zellij support, a stripped-down no-AI build—these poured in across channels. The latest tagged release arrived April 29, labeled v0.2026.04.29.08.56.stable_00.
The Money Behind the Experiment
Warp has raised approximately $73 million since its 2020 founding. A $6 million seed led by GV. A $17 million Series A led by Dylan Field of Figma. A $50 million Series B in June 2023, with Sequoia Capital at the helm.
Lloyd, formerly a principal engineer on Google Docs, shipped Warp's Windows version in February 2025. The tool now runs across major desktop platforms. The open-source move positions the terminal as the visible, community-facing layer while keeping the orchestration platform—where Warp sees the revenue—proprietary.
It's a bet. Will developers embrace agent-mediated contributions, or will they fork the code and strip out the AI scaffolding entirely? Both outcomes seem plausible.
For now, the dashboard ticks. Oz reviews. The community files issues. Open source, yes—but filtered through a layer of automation that has drawn mixed reactions from early contributors.
It's not quite like anything that came before it. Whether that's an advantage or a complication, the next few months will likely tell.
