Gemini CLI End-of-Life and the Antigravity CLI (agy): What Actually Changes
Gemini CLI End-of-Life and Its Successor Antigravity CLI (agy): What Actually Changes
The free-to-use Gemini CLI reaches end-of-life on June 18, 2026. Its replacement is Antigravity CLI — invoked as agy on the command line. Two points matter most: "Antigravity CLI" is not a standalone tool — it is one component of a larger agent-first development platform — and the project has shifted from open-source to closed-source. This post cross-references official announcements and major press coverage to lay out the full picture: what Antigravity is, how it differs from Gemini CLI, and how to install it.
Antigravity Is a Platform, Not Just a CLI
The most common misconception needs addressing first. Antigravity's core product is an agent-first development platform — the CLI is just one interface within it. When Antigravity was first unveiled alongside Gemini 3 on November 18, 2025, it debuted as a desktop GUI IDE. In other words, "Antigravity = GUI IDE" came first; the terminal CLI (agy) was added later as one component of the broader platform.
"Antigravity 2.0," announced at Google I/O on May 19, 2026, bundles the following into a single package. The key point is that all components share the same agent harness — the agentic specialization is a platform-wide design decision, not a property of any single interface.
▶ Desktop IDE (GUI mission control)
▶ CLI (agy) — terminal integration and automation
▶ SDK — programmatic integration
▶ Managed Agents API — managed agent tier
▶ Enterprise deployment path
flowchart TD
A([Antigravity 2.0
Agent Platform]) --> B[Desktop IDE
GUI Mission Control]
A --> C[CLI · agy
Terminal Automation]
A --> D[SDK + API
Managed Agents]
B --> E([Shared Agent Core])
C --> E
D --> E
style A fill:#e8f8f5,stroke:#16a085,color:#117a65
style B fill:#eaf2f8,stroke:#2980b9
style C fill:#fef9e7,stroke:#f39c12
style D fill:#f4ecf7,stroke:#8e44ad
style E fill:#16a085,stroke:#117a65,color:#ffffff
🔗 Diagram summary: The Antigravity 2.0 platform exposes three interfaces — GUI IDE, CLI (agy), and SDK/API — but all three share the same agent core. "Agentic specialization" is therefore a platform-wide design philosophy, not a distinction between one interface and another.
Key Dates at a Glance
Much of the confusion stems from two separate announcement timelines: the IDE launched in November 2025, while the CLI and the end-of-life notice both arrived at the May 2026 I/O.
(alongside Gemini 3)
(Google I/O 2026)
stops accepting requests
Starting June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extension will stop processing requests for Google AI Pro/Ultra subscribers and free individual users. The announcement itself came roughly one month earlier, on May 19.
GUI vs. CLI — What Actually Differs
The phrase "more specialized for agentic behavior" does not distinguish the GUI from the CLI. As shown above, both share the same agent core. The real differences are interface form and which workloads each suits best.
| Dimension | 🖥️ GUI (Desktop IDE) | ⌨️ CLI (agy) |
|---|---|---|
| Interface | Editor View (code editor) + Agent Manager (control dashboard) | TUI: conversation panel + > prompt + status bar |
| Strengths | Visually create and monitor multiple agents; validate results in a browser | Direct integration with git, compilers, and CI/CD; scriptable and automatable |
| Best for | Visual orchestration of complex async workflows | Immediate automation without leaving your existing terminal workflow |
The GUI's Agent Manager acts as a mission-control panel, giving you a single view of multiple agents working asynchronously across different workspaces. The CLI delivers the same capability in a lightweight, scriptable form that stays inside the terminal. The choice comes down to what you are trying to accomplish — this is a context question, not a capability question.
flowchart TD
A([Need a terminal AI]) --> B{Is visual
oversight important?}
B -->|YES| C[Desktop IDE
Use Agent Manager]
B -->|NO| D[CLI · agy
Script automation]
style A fill:#3498db,stroke:#2980b9,color:#ffffff
style B fill:#fef9e7,stroke:#f39c12
style C fill:#eaf2f8,stroke:#2980b9,color:#2471a3
style D fill:#eafaf1,stroke:#27ae60,color:#1e8449
🔁 Diagram summary: If you need to visually manage and verify multiple agents, use the desktop IDE. If you want script-level automation that fits into an existing terminal or CI/CD pipeline, use the CLI (agy). Since both run the same core, this is a workflow fit question, not a feature question.
Who Gets Hit by the Gemini CLI Shutdown
The impact is not evenly distributed. Individual and free-tier users bear the brunt; paid enterprise customers are effectively unaffected.
🔴 Affected users
Google AI Pro/Ultra subscribers and free individual users who rely on Gemini CLI or the Gemini Code Assist IDE extension. Requests stop being processed after June 18.
🟢 Unaffected users
Organizations on Gemini Code Assist Standard or Enterprise licenses, and organizations using Gemini Code Assist for GitHub via Google Cloud. Paid enterprise environments continue as-is.
If you are an individual developer, the practical advice is clear: migrate to agy before June 18, or identify an alternative terminal AI tool ahead of time.
Implementation Language and Supported Models — Correcting Common Misinformation
Gemini CLI was TypeScript, not Python
Some sources describe Gemini CLI as "Python-based," which is incorrect. The official repository (google-gemini/gemini-cli) is written in Node.js (≥20) + TypeScript, uses React + Ink for terminal rendering, and was released under the Apache 2.0 open-source license. The implementation language of its successor agy has not been officially confirmed — claims about Go (or any other language) should be treated as unverified.
Model optionality: not locked to a single backend
One of the Antigravity platform's stated strengths is that it is not tied to a single model. This matters in practice: you can swap backends mid-session with /model, routing latency-sensitive tasks to a lighter model and complex reasoning to a heavier one. Confirmed supported models include the following.
Gemini 3 Pro
Generous rate limits (Google's own model)
Claude Sonnet 4.5
Anthropic model
GPT-OSS
OpenAI open-weight model
⚠️ Common misinformation: Claims that "Gemini 3.5 Flash is the default engine" are not backed by official sources. What is confirmed is the Gemini 3 Pro family. Similarly, curl ... | bash-style install commands circulating online have not been officially verified — treat them with caution.
Why Google Closed the Open-Source Project
The decision to retire open-source Gemini CLI in favor of the closed-source agy reflects both technical and business motivations.
① Technical motivation. Supporting multi-agent parallel collaboration and asynchronous task execution — well beyond a single agent running tasks sequentially — required a new agent harness architecture. Antigravity redesigns this as a unified core shared by the IDE, CLI, and SDK, rather than bolting async orchestration onto an existing single-agent CLI.
② Business motivation (the controversial part). Unlike Gemini CLI, agy is not open-source. Gemini CLI had accumulated roughly 6,000 community contributions under Apache 2.0. Critics argue that Google absorbed that community effort, channeled the value into an enterprise product, and replaced it with a proprietary successor.
🧠 Developer communities (Hacker News, FOSS Force, and others) have characterized this pattern as a "bait-and-switch": attract 6,000 contributions under an open-source license, then lock the resulting work behind a proprietary product. The concern is that this precedent could erode community trust in future Google open-source initiatives — making contributors hesitant to invest effort in projects that may later be closed.
At a more fundamental level, the shift represents a change in the developer's role. Instead of typing low-level commands directly, the developer becomes someone who delegates goals to agents and then validates and manages the results — a workflow that favors a platform architecture over a simple CLI wrapper.
macOS Installation and Basic Usage
1) Install via Homebrew (verified path)
The confirmed installation path on macOS is Homebrew (macOS 12 or later required).
The installed binary is named agy — not antigravity — and lands in ~/.local/bin/. On first launch, a browser window opens automatically for Google OAuth login; credentials are cached in the macOS Keychain. Verify the install by checking the version number printed at startup.
2) Basic usage — TUI and core commands
Running agy from a project directory opens a full-screen TUI: a scrollable conversation panel, a > prompt, and a status bar showing the active model, token count, and any running sub-agents.
| Feature | Description |
|---|---|
| @ file reference | Pull files into context without copy-paste: @src/ (directory) or @**/*.ts (glob) |
| /goal | Run the task to completion without intermediate approval checkpoints (omitting it prompts you at each step) |
| /model | Switch models mid-session |
| /btw | Handle a side task in a temporary overlay without interrupting a running background operation |
| Other commands | /schedule /resume /rewind /permissions /skills /mcp /tasks /logout |
Customization is a first-class feature. An AGENTS.md plain-text file in the project root defines behavioral rules for the agent — analogous to a system prompt scoped to your repo. Markdown files under .agents/skills/ (e.g., lint.md) are automatically registered as slash commands — callable as /lint — making reusable, version-controlled skill definitions easy to maintain across a team.
Summary
agy is not a simple upgrade of Gemini CLI. It represents a simultaneous shift along three axes.
✓ Open-source → closed-source (the root of community trust concerns)
✓ Single CLI → multi-interface platform (IDE + CLI + SDK sharing one core)
✓ Free individual → paid enterprise focus (individual users absorb the impact)
Functionally, the GUI and CLI share the same agent core, so "agentic specialization" is a platform-wide design philosophy, not a distinguishing property of one interface over the other. The selection criterion is straightforward: use the GUI for visual mission control of multiple agents; use the CLI to embed agent capabilities into a terminal or automation pipeline.
Areas still awaiting official confirmation: the implementation language of agy (Go claims are unverified), the exact default model name, and the official install command for paths other than Homebrew. With June 18 approaching, the practical recommendation for individual developers is to install agy and migrate your workflow before the deadline.
References
→ Google Developers Blog — Introducing the Antigravity Platform
→ Google Developers Blog — Transitioning Gemini CLI to Antigravity CLI
→ The Register — Bye-bye Gemini CLI
→ gemini-cli GitHub repository
→ Homebrew Formulae — antigravity-cli
→ DataCamp — Antigravity CLI Tutorial
This post cross-references official announcements and major technical press coverage. Items that remain unverified are flagged inline. Tool pricing, timelines, and features are subject to change at the provider's discretion — consult official documentation before installing or migrating.
Research, organize, and verify — that's the workflow behind every post on this blog.
This post is based on publicly available data and cited sources. Last updated: June 8, 2026
댓글
댓글 쓰기