Warp’s terminal client is now open-source under AGPL, using an agent-first contribution model where OpenAI’s GPT models power coding via their Oz orchestration platform.
Key Takeaways
Repo at github.com/warpdotdev/warp under AGPL; contribution model has agents handle coding and testing while human contributors focus on direction and verification.
OpenAI is the founding sponsor; all agentic workflows run through Oz, Warp’s cloud agent orchestration platform, on GPT models.
New open-source model support ships today: Kimi, MiniMax, Qwen, plus an “auto (open)” router that selects the best open model per task.
New settings file enables programmatic config control and device portability; UI is now customizable from a bare terminal to a full ADE with diff view and file tree.
Framed as a direct competitive move: community-supervised agent scale replaces internal headcount against highly funded, closed-source rivals.
Hacker News Comment Review
Privacy skepticism is the loudest thread: commenters question whether Warp initiates network connections without explicit user consent, citing the old mandatory account requirement as evidence of prior value misalignment.
The “human bottleneck” framing is contested: the more cynical read is that open-sourcing offloads compute and review costs onto the community without surrendering meaningful creative control to contributors.
Demand for a terminal-only fork stripped of AI and cloud features is strong and vocal; the absence of the full commit history blocks any clean branch point from before the agentic pivot.
Notable Comments
@SuperV1234: Account disabled on first use after asking the agent to open tabs per build folder; a ToS ban for a routine request is a concrete credibility problem for the open contribution model.
