Gemini CLI Developer Tutorial: Must-Know Info Before June 18

Google’s open-source terminal coding agent, Gemini CLI, stopped serving requests on June 18, 2026, for free-tier, Google AI Pro, and Ultra users. If you’ve been using the gemini command in your terminal, CI pipelines, or automation scripts, it’s already broken. The replacement—Antigravity CLI—is a fundamentally different tool with a closed-source architecture, a new command surface, and a quota model that represents a roughly 98% reduction in free usage.

This isn’t a typical version upgrade. It’s a forced migration that exposes a pattern I’ve been tracking across the AI tooling ecosystem: open-source community labor gets harvested to validate a product, then the infrastructure gets closed to monetize enterprise users exclusively. The Apache 2.0 license on Gemini CLI’s repository remains intact, but without Google’s auth endpoints and API backend, the code is effectively a skeleton.

What Actually Happened to Gemini CLI

On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions stopped serving requests for consumer tiers—free, Google AI Pro, and Ultra subscribers. Enterprise Gemini Code Assist Standard and Enterprise license holders are explicitly exempt and retain uninterrupted access. The consumer version of Gemini Code Assist for GitHub blocked new organizational installations the same day, with full shutdown of existing code review activity scheduled for approximately July 17, 2026.

The transition was announced at Google I/O on May 19, 2026, providing roughly a 30-day migration window. Google framed the change as a response to developer workflows that “outgrew those early days of 2025” and now require multi-agent orchestration. The community response was less charitable—contributors like Andrea Alberti, who had a 27-commit pull request merged the same day as the shutdown announcement, asked directly whether they “essentially worked for free on a code base that will only be used in enterprises.”

The technical rationale has merit: Gemini CLI was built as a TypeScript application running a single-agent ReAct loop, which doesn’t scale to the parallel, asynchronous workflows that agentic development demands. But the business logic is what stings. Google accepted over 6,000 merged pull requests from external contributors to the Apache 2.0 project, which accumulated over 100,000 GitHub stars, then systematically transitioned to a closed-source successor that excludes the original contributors.

Antigravity CLI: The Closed-Source Successor

Antigravity CLI (agy) is a ground-up Go rewrite, not a fork of the original TypeScript stack. It’s designed for asynchronous multi-agent workflows—a manager agent decomposes goals into subtasks and spawns dynamic subagents in parallel, each with isolated context windows and tool sets. Google describes it as “snappier” and built for “complex tasks in the background.”

The catch: there’s no 1:1 feature parity at launch. Google explicitly states this in their transition announcement. Core features like Agent Skills, Hooks, Subagents, and Extensions (now called plugins) transfer, but the implementation details differ in ways that break existing automation.

The free tier economics collapsed. Gemini CLI offered 1,000 requests per day on a personal Google account. Antigravity CLI’s free tier provides roughly 20 requests per week—a reduction of approximately 98%. Heavy users on the Antigravity issues tracker report exhausting the weekly quota in a handful of prompts.

The Migration Path: What You Actually Need to Do

If you’re migrating from Gemini CLI to Antigravity CLI, the official path includes several steps that sound simple but hide gotchas:

1. Install Antigravity CLI from antigravity.google
2. Run agy plugin import gemini to bring over existing extensions
3. Rename GEMINI.md to AGENTS.md
4. Move .gemini/skills/ to .agents/skills/
5. Update scripts referencing gemini to use agy

The automated migration script does not update CI configurations, cron jobs, or shell aliases. You must audit those manually. This is where teams get burned—a shell script calling gemini in a GitHub Actions workflow fails silently or returns a hard 403 mid-session.

The MCP config change is another silent breaker. Antigravity CLI uses a separate mcp_config.json file for MCP configuration, whereas Gemini CLI used inline configuration in settings.json. Pipelines that loaded MCP servers the old way fail without an obvious error message.

The Open-Source Capture Pattern

What happened to Gemini CLI isn’t unique—it’s just the most visible recent example of an ecosystem capture pattern where open-source community labor gets absorbed into a closed enterprise product. The Apache 2.0 repository remains publicly accessible with continued releases (v0.45.0, v0.47.0, v0.49.0) after the shutdown announcement. But without Google’s backend infrastructure, the code can’t serve requests for affected tiers.

Antigravity CLI’s public GitHub repository contains only a changelog, readme, and GIF—no application code. The open-source license governs the code, but it says nothing about API access, model quotas, authentication systems, or the cloud infrastructure that determines whether the tool actually works.

This is the infrastructure control problem that open-source licensing can’t solve. When the value lives in the cloud layer—the auth endpoints, the model serving, the quota management—the code being open is theater. Contributors who signed a CLA gave Google the rights to use their labor; the license gave Google no obligation to keep the service open.

Should You Migrate or Move On?

The decision framework depends on your constraints:

Migrate to Antigravity CLI if: You’re already in the Google ecosystem, your workflows are interactive rather than automated, and you can live with weekly quota limits. The async multi-agent architecture is genuinely better for large-scale refactors and parallel research tasks.

Claim the enterprise exemption if: Your organization has Gemini Code Assist Standard or Enterprise licenses, or you’re using paid Gemini API keys. These tiers retain uninterrupted access to both Gemini CLI and Antigravity CLI.

Move off Google entirely if: You rely on CI/CD automation, you need predictable quotas, or you don’t want to bet on a vendor that just sunset a community-built tool. For open-source control, Aider runs locally with any model provider.

The broader lesson: when evaluating AI coding tools, look past the open-source license and ask who controls the backend. If the value is in the cloud layer, the code being open is a feature, not a guarantee.

Feature Comparison: Gemini CLI vs. Antigravity CLI

Dimension	Gemini CLI	Antigravity CLI
Binary name	gemini	agy
Language	TypeScript	Go
Source model	Apache 2.0 open-source	Closed-source
Architecture	Single-agent ReAct loop	Async multi-agent orchestration
Free tier quota	1,000 requests/day	~20 requests/week
MCP config	Inline in settings.json	Separate mcp_config.json
Skills directory	.gemini/skills/	.agents/skills/
Extensions	Extensions API	Plugins API
Feature parity	Baseline	No 1:1 parity at launch

The tools that win long-term are the ones that integrate transparently into existing workflows rather than demanding workflow rewrites. Google’s transition prioritizes enterprise monetization over the developer ecosystem that validated the product. Whether that’s strategically rational or community exploitation depends on your tolerance for vendor lock-in—and whether you were one of the 6,000 contributors who built the thing that got taken away.