Gemini CLI Ends June 18: Everything You Need to Know Before Migrating to Google's Antigravity CLI

On June 18, 2026 — two days from today — Google will shut down the hosted response path for Gemini CLI across all consumer account tiers. That means every command-line workflow, automation script, Git hook, GitHub Actions job, and scheduled cron task that shells out to `gemini` will return nothing and break silently unless migrated before the deadline. The replacement is Antigravity CLI, the terminal surface of Google's newly unified Antigravity 2.0 platform. The migration itself is straightforward in simple cases, but four behavioral differences between the old and new tools will silently break teams that treat it as a rename. Quota mechanics also change in ways that hit power users hard. Here is what every developer who has ever typed `gemini` in a terminal needs to understand before Thursday morning.

What Google Antigravity 2.0 Actually Is

Google announced Antigravity 2.0 at Google I/O on May 19, 2026. It is not a product upgrade — it is a platform consolidation. Antigravity 2.0 replaces Gemini CLI, the original Gemini Code Assist IDE extensions, and the standalone Gemini Code Assist desktop client with a single unified agent-first development platform that ships five surfaces at once: a redesigned standalone desktop application built around multi-agent orchestration, a new Antigravity CLI (`agy`) written in Go and focused on terminal workflows, an Antigravity SDK for building custom agents programmatically against the same runtime Google uses internally, a Managed Agents tier inside the Gemini API for teams that want orchestration-as-a-service, and an enterprise deployment path through the Gemini Enterprise Agent Platform. The core architectural shift is from a tool that responds to single prompts to a platform that coordinates multiple specialized subagents working on different parts of a problem in parallel — up to five simultaneous agents in the desktop app. For developers using Gemini CLI as a simple command-line assistant, this is more platform than they asked for. But the old CLI is being retired regardless, and the June 18 deadline is hard.

Who Gets Cut Off on June 18

The shutdown on June 18 affects all Google consumer tiers: free accounts, Google AI Pro subscribers at $19.99 per month, and Google AI Ultra subscribers at $29.99 per month. After the deadline, Gemini CLI will continue to accept commands but will return no AI-generated responses — it stops serving requests. Enterprise customers are exempted if they hold an active Gemini Code Assist Standard or Enterprise license, a Gemini Code Assist for GitHub license through Google Cloud, or a paid Gemini API key or Gemini Enterprise Agent Platform subscription. For enterprise customers on one of those qualifying plans, the old Gemini CLI surface continues to work while Google manages their migration timeline separately. For everyone else — individual developers, small teams, open-source contributors, freelancers — June 18 is the hard cutoff.

The Four Breaking Behaviors That Will Catch You Off Guard

Antigravity CLI is described as a near-drop-in replacement for Gemini CLI, and for interactive use that description is accurate. For automation, four behavioral differences break scripts that are not updated beyond a binary rename. First, the default model changes. Gemini CLI defaulted to Gemini Ultra. Antigravity CLI defaults to Gemini 2.5 Flash. Any script that relies on the implicit model-tier assumption without a `--model` flag will now get Flash-quality responses instead of Ultra-quality responses — a significant quality regression for complex tasks that relied on the higher tier implicitly. Second, the streaming output format changes. Gemini CLI streamed tokens progressively to stdout. Antigravity CLI buffers the full response and writes it once, then exits. Any script that reads and processes streaming output line-by-line from the Gemini CLI subprocess — log processors, interactive pipelines, real-time display UIs — will receive the full response at once at the end, breaking anything that expected incremental data. Third, the state directory moves. Gemini CLI stored its conversation history and session state in `~/.gemini`. Antigravity CLI stores it in `~/.antigravity`. Scripts that reference the state directory path explicitly, or that clean, backup, or inspect session files by path, will silently stop finding what they expect. Fourth, exit codes change semantics. Gemini CLI returned exit code 0 for any response that produced output, including model refusals and error messages embedded in the response text. Antigravity CLI returns non-zero exit codes for model errors and content policy blocks. Scripts that use the Gemini CLI exit code as an indicator of successful generation — rather than a network call indicator — will now behave differently on model refusals.

The Quota Regression: From 1,000 Daily Requests to a Weekly Cap

Beyond the breaking behavioral differences, the quota model change is the issue most likely to affect developers who migrate successfully on the first try. Gemini CLI operated on a daily request cap of 1,000 requests per day, which reset every 24 hours. Antigravity CLI operates on a weekly compute-based cap, with the cap measured in compute units rather than raw request count. The total weekly allocation varies by tier, but the structure change — daily reset versus weekly reset — has a compounding effect on workflows that depend on predictable refresh cycles. Developers who run nightly batch jobs, daily test suites, or scheduled summarization pipelines found that Gemini CLI's daily reset aligned naturally with their automation cadence. With Antigravity CLI, a single heavy day of automated use can exhaust the weekly compute cap, triggering a multi-day cooldown that halts automation until the next weekly reset. Several developers testing the migration before the June 18 deadline reported exhausting their weekly cap within the first 18 to 24 hours of automated use, then facing a four-to-six-day wait before their quota recovered. For teams that used Gemini CLI heavily in CI/CD pipelines — running code review bots, automated documentation generation, or test coverage analysis on every commit — the quota model change may require a fundamental rethink of how often those workflows run.

What Breaks in Your CI/CD Pipeline

The practical blast radius of the June 18 shutdown is widest for developers who automated Gemini CLI into their developer toolchain. Any GitHub Actions workflow that shells out to `gemini` will fail on June 18 — the command will run but return no model response. Git pre-commit and pre-push hooks that invoke Gemini CLI for automated code review or commit message generation will silently produce empty output or break the hook chain depending on how exit codes are handled. Cron jobs running on VMs or containers that call `gemini` for daily summarization, log analysis, or data processing will produce empty results. Deploy scripts that use Gemini CLI to generate change summaries or release notes will need to be updated. The migration playbook for each of these cases involves three changes: replace the `gemini` binary reference with `agy`, add an explicit `--model gemini-ultra` flag to avoid the default model regression, and audit any stdout handling for the streaming-to-buffered output format change. For GitHub Actions specifically, any step that uses streaming-aware output processing — piping to `tee`, using while-read loops on the subprocess output — needs to be rewritten to handle buffered output.

The Migration Steps

For developers who use Gemini CLI interactively rather than in automation, migration is a 15-minute operation: install the Antigravity CLI with `brew install antigravity` on macOS, `winget install Google.Antigravity` on Windows, or the Linux `.deb` or `.rpm` package from the Antigravity release page, authenticate with `agy auth login` using the same Google account, verify with `agy --version`, and you are done. For developers who have Gemini CLI baked into automation, the migration requires a systematic audit. Search your repository for every invocation of the `gemini` binary — in shell scripts, Makefiles, GitHub Actions YAML, Dockerfiles, and CI configuration. For each invocation, replace `gemini` with `agy`, add `--model gemini-ultra` if the task requires Ultra-level reasoning, review any stdout processing for the buffered output format change, update any path references from `~/.gemini` to `~/.antigravity`, and add exit code handling for the new non-zero error semantics. If your team depends on daily quota resets for automation cadence, evaluate whether to reduce automation frequency to stay within the weekly compute cap or move to paid API access through the Gemini API or Gemini Enterprise Agent Platform, which provides per-request billing rather than weekly caps.

Why This Matters Beyond the Migration

The Gemini CLI shutdown is the most immediate issue, but it is part of a larger strategic move that has implications beyond a single tool migration. Google is retiring a popular open-source tool — Gemini CLI accumulated over 100,000 GitHub stars — and replacing it with a closed-source Go binary whose full behavior is not publicly inspectable. The Antigravity SDK is open-source, but the CLI itself is not. For teams that embedded Gemini CLI into their toolchain based on its open-source licensing, this is a trust and auditability regression. The platform consolidation also signals Google's strategic direction: the company is betting that developers will manage AI agents through multi-agent orchestration platforms rather than single-model terminals. The Antigravity 2.0 architecture — with its five-surface parallel agent model, SDK, and enterprise tier — is designed for teams building AI-native development workflows, not for individual developers running one-off CLI queries. The June 18 migration is the moment Google ends the era of Gemini CLI as a simple AI terminal and begins the era of Antigravity as a developer agent platform. For individual developers who found Gemini CLI useful as a lightweight tool, the new platform is more than they need. For engineering teams building automated AI workflows at scale, the new architecture may be exactly what they have been waiting for.

Bottom Line

Gemini CLI stops working on June 18. If you use it interactively, migrate to `agy` today — it takes 15 minutes. If you have it in automation, perform the full audit: binary rename, explicit model flag, streaming output handler update, state directory path update, and exit code handling review. Check your weekly quota before running your heaviest automation workflows. The bigger picture is that Google is retiring its lightweight terminal AI tool in favor of a multi-agent orchestration platform. Whether that is an upgrade or an overreach depends on whether your use case is a one-person AI terminal or an engineering team building agentic development workflows. Either way, the deadline is not moving.