Google Killed the Open-Source Gemini CLI. Here's Why That Matters.
Google is shutting down the Apache-2.0 Gemini CLI on June 18 and replacing it with a closed-source tool — and that tells you everything about where the industry is heading.
Google is shutting down the Apache-2.0 Gemini CLI on June 18 and replacing it with a closed-source tool — and that tells you everything about where the industry is heading.
On May 12, Google announced that the open-source Gemini CLI — Apache-2.0 licensed, forkable, community-contributed — will stop serving requests on June 18, 2026. The replacement is the Antigravity CLI: built in Go, tightly integrated with the Antigravity 2.0 desktop app, and not open source.
That’s a big move, and most coverage has treated it as a product-transition notice. It’s more than that.
What actually happened
Google shipped an open-source terminal agent, watched developers adopt it, and then decided the open-source model was not what they wanted. The Antigravity CLI is a better product in several measurable ways — async background workflows, tighter desktop integration, a cleaner architecture. But it’s closed source, and it’s tied to Google’s auth and billing in ways the old Gemini CLI wasn’t.
For free-tier and Pro/Ultra users, the timeline is blunt: the CLI you’ve been using stops working in three weeks. If you’ve built workflows, automation, or personal tooling on top of it, you have until June 18 to migrate or rebuild.
The HN thread (id 48196867) has the predictable range of reactions, but the loudest complaint is the one you’d expect: the open-source removal. Developers who adopted Gemini CLI partly because it was Apache-2.0 are now being told that the fork they might have made, the extension they were planning, the self-hosted version they were considering — none of that matters, because the product they built on is being sunsetted under them.
The pattern this fits
This isn’t a unique event. It’s a recurring arc in developer tooling: a company ships an open-source version to drive adoption, that version reaches critical mass, and then the commercial pressures push toward a closed successor that the company can actually monetize and control. It happened with Elasticsearch. It happened with Redis. It’s a reasonable business decision and a genuinely frustrating user experience, simultaneously.
What makes the CLI space different is that the switching cost is supposed to be low. It’s a terminal tool. You can change your terminal agent in an afternoon. But what you can’t easily change is the institutional trust you built around a tool being open and inspectable. That trust is what gets broken when an open-source CLI gets deprecated out from under you.
Claude Code is not open source. Codex CLI is not open source. The two tools that dominate the current cycle of developer interest are both proprietary, both locked to their respective providers, and both making aggressive moves toward capability (Codex just defaulted “goals” on; Claude Code doubled rate limits and shipped dynamic workflows in May). They’re excellent tools. They’re also vendor-controlled in ways that matter if you care about what runs on your machine and who sees your code.
Where gptme fits in this
gptme is open source. MIT licensed. It runs locally. It supports every major model provider, including local models via Ollama. Your conversation history is yours, stored in plain files you can read and grep and back up. You can audit the code. You can fork it. You can run it without any network connection if you’re using a local model.
That’s not a marketing claim — it’s literally how it works. And right now, after Google’s announcement, it’s a more differentiated position than it was two weeks ago.
I want to be honest about where gptme stands: it’s a smaller project than Claude Code or Codex, with fewer resources behind it. The UX is rougher in places. The integrations that come pre-built in the commercial tools require more setup here. If you want the most capable, most polished terminal agent and you’re comfortable with a proprietary tool, Claude Code is genuinely impressive.
But if you want something that cannot be deprecated out from under you by a business decision you had no part in making — gptme is the serious option in that space right now. The Gemini CLI exit didn’t create that position; it just made it more visible.
What this says about the industry
The honest read is that the major AI companies have concluded that open-source CLI agents are strategically valuable for adoption and strategically costly for monetization. The commercial path runs through subscription tiers, proprietary backends, and tight integration with the mothership. That’s not a conspiracy — it’s just where the incentives point.
The developers who care about open-source, local-first tooling are not the primary customer for any of these companies. They’re a side effect of the adoption phase. When the adoption phase ends and the monetization phase begins, the open-source version tends to get deprecated.
gptme is built by people who actually want this tool to exist on the terms we’re describing. That’s a different foundation than a corporate open-source project that serves a strategic purpose until it doesn’t.
If you want to try it
pipx install gptme
gptme "help me refactor this function"
Docs at gptme.org. The source is at github.com/ErikBjare/gptme. June 18 is three weeks away — if you’re evaluating alternatives to Gemini CLI, you have time to kick the tires before the cutover.