Google killed Gemini CLI on June 18 — replacing a 104,000-star Apache 2.0 open-source tool with a closed-source Go binary called Antigravity CLI. The community’s response was about as warm as you’d expect: 143 thumbs-down reactions on the official GitHub announcement thread versus four cheers. A month in, developers are still sorting out whether the trade is worth it.
The Open Source Problem #
Gemini CLI launched in June 2025 under Apache 2.0 with an active contributor community that merged 6,000+ pull requests. Antigravity CLI is proprietary. You can’t audit it, fork it, or self-host it. The closed-source decision hit a raw nerve not just because of the license change, but because of the implied sequence: Google benefited from thousands of hours of community contribution, then shipped those improvements into a product the community can no longer touch.
The Register didn’t mince words: “Bye-bye, Gemini CLI; Google’s gone and swapped you for a closed-source AI.” That about covers it. If you were contributing to Gemini CLI’s development, those contributions are now locked inside someone else’s proprietary product.
What Actually Changed #
Beyond the license, there’s a real list of differences that affect daily usage.
Rate limits took a serious hit. Gemini CLI’s free tier allowed roughly 1,000 requests per day. Antigravity CLI’s free tier gives you approximately 20 agent requests per day — and usage is now compute-based, meaning complex multi-step tasks eat through quota faster than simple prompts. The quota refreshes every five hours. For hobbyists and small teams running lightweight automation, this alone may be a dealbreaker.
The tool itself is architecturally different. Gemini CLI was a Node.js terminal assistant — a single agent responding to your prompts. Antigravity CLI is a multi-agent orchestration engine written in Go. The primary agent can define and spawn subagents mid-task, run them in parallel, and keep the main context window clean. It also shares a “Shared Agent Harness” with the Antigravity desktop IDE, so improvements to the core agent engine roll out across CLI, desktop, and SDK simultaneously.
The Go rewrite makes the binary noticeably faster. And Antigravity CLI is multi-model — you can route tasks to Claude, GPT-OSS models, or Gemini, not just Google’s own stack.
Migration: What You Need to Do #
The migration itself is straightforward. Install with:
curl -fsSL https://antigravity.google/cli/install.sh | bash
On first launch, agy
prompts you to import your existing ~/.gemini/
config — MCP servers, allowed commands, keybindings, themes. If the prompt doesn’t appear, run agy plugin import gemini
manually. That handles most of the transition.
The one thing that doesn’t migrate automatically: CI and cron pipelines that invoke the gemini
binary directly. Those need manual updates to call agy
instead. Check your scripts, GitHub Actions workflows, and scheduled jobs before assuming the migration is complete. Skills also need to move from .gemini/skills/
to .agents/skills/
.
See Google’s official transition guide and the DataCamp walkthrough for the full details.
If You Don’t Want to Follow Google #
The living open-source successor to Gemini CLI is Qwen Code — Alibaba’s fork of the Gemini CLI codebase, Apache 2.0 licensed and model-agnostic. It works with Claude, GPT, Gemini, and local models via Ollama. It’s sitting at around 25,000 GitHub stars, which for a tool that launched as a direct response to Google’s closure is fast traction. If open-source continuity matters more than Google’s multi-agent features, Qwen Code is the right move.
For developers who want to stay in the Google ecosystem but need to understand what they’re giving up, this analysis of the open-source trust implications is worth reading before you commit.
The Bottom Line #
The multi-agent upgrade in Antigravity CLI is real. Parallel subagents, a shared harness, Go-native speed, and multi-model support are meaningful improvements over what Gemini CLI offered. But Google made a bad-faith move with the open-source community — using years of contributor work to build a better closed product. Enterprise teams with Google Cloud spend who want orchestrated agents will probably find the migration worth it. Developers who value open-source trust and don’t need 20-requests-per-day counting should look at Qwen Code instead.
The 143:4 ratio on the GitHub thread tells you which group is larger.