Gemini CLI Is Dead: Complete Migration Guide to Antigravity CLI (2026)

Sometime after June 18, 2026, a lot of developers ran a script that had worked fine for months and got back a wall of HTTP 410 Gone errors. If that's you, the short version is: it's not your code. Google shut down Gemini CLI for free, Pro, and Ultra accounts on that exact date, and replaced it with a new tool called Antigravity CLI, invoked as agy.

I run AI automation and agent workflows for clients, so a chunk of my own tooling was directly affected by this. This guide is the migration path I actually used, cross-checked against Google's own transition docs and the Antigravity CLI GitHub repo — every command below is something you can run today, not a summary of a press release.

Why Gemini CLI Stopped Working

At Google I/O on May 19, 2026, Google announced it was consolidating its developer tooling under one brand — Antigravity — and retiring the standalone Gemini CLI and Gemini Code Assist IDE extensions. The stated reason: developer workflows had "outgrown those early days of 2025" and now require multiple agents working together, communicating, and splitting up complex tasks — something the original Node.js-based Gemini CLI wasn't architected for.

The technical shift is real, not just a rebrand:

Node.js → Go. Antigravity CLI is a compiled Go binary, giving it meaningfully faster startup and lower memory use than the old Node.js-based Gemini CLI, especially on large codebases.
Synchronous → asynchronous agents. Antigravity supports background agent workflows — you can kick off a project-wide refactor and keep using your terminal instead of it locking up.
Fragmented → unified harness. The same agent engine now powers the CLI, the Antigravity desktop IDE, and IDE extensions, so features land across all three simultaneously instead of the CLI lagging behind.

If you're seeing this right now

Your gemini binary still exists on disk and will still launch — it just can't reach Google's servers anymore for free/Pro/Ultra accounts, so every request returns HTTP 410 Gone. Nothing is wrong with your local install; the endpoint itself was retired.

Who's Actually Affected

This is the first thing to check, because not everyone needs to migrate immediately:

Account Type	Status	Action Needed
Free tier	Cut off June 18, 2026	Migrate now
Google AI Pro / Ultra	Cut off June 18, 2026	Migrate now
Gemini Code Assist for individuals (free)	Cut off June 18, 2026	Migrate now
Gemini Code Assist Standard / Enterprise	Unaffected	None — keep using Gemini CLI, or evaluate agy at your own pace
Vertex AI / paid Gemini API key	Unaffected	None required

To check which category you're in, run gemini --version and look at the account type in the output header. If you're unsure, the safest assumption is that if you were never issued a Google Cloud enterprise license or Workspace account, you're on the affected list.

What's Actually Changing (Not Just a Rename)

The core CLI syntax stays close to what you already know, but a few structural things move, and these are exactly the parts that break scripts silently if you don't catch them:

Authentication — Antigravity CLI replaces manual API key management with a browser-based Google OAuth flow by default, storing tokens in your OS's native keyring instead of a flat config file.
Environment variable — GEMINI_API_KEY becomes AV_API_KEY for programmatic/headless use.
State directory — session data moves from ~/.gemini/brain/ to ~/.gemini/antigravity-cli/brain/. Checkpointing works the same; only the path changed.
MCP config format — MCP server definitions move out of settings.json into a dedicated mcp_config.json, and the url field is renamed to serverUrl.
Exit codes and streaming — Antigravity CLI defaults to SSE-based streaming and introduces new non-zero exit codes that older scripts checking for specific Gemini CLI exit statuses won't recognize.

Step 1: Install Antigravity CLI

Installing agy does not touch or break your existing gemini installation — they're separate binaries, so you can install Antigravity CLI today and migrate at your own pace before the deadline fully affects your workflow.

# macOS / Linux — official install script
curl -fsSL https://antigravity.google/cli/install.sh | bash

# macOS / Linux — Homebrew
brew install antigravity-cli

# Any platform — npm
npm install -g @google/antigravity-cli

The install script auto-detects your OS and architecture, downloads the matching build, verifies the file's SHA-512 checksum, and drops the agy binary into ~/.local/bin/agy.

Windows note

After installing, you'll very likely see a warning that the install folder isn't on your PATH yet — this is the single most common reason people think the install failed when it actually didn't. Add the install directory to your PATH and open a new terminal session before troubleshooting further.

Step 2: Authenticate

Run agy with no arguments on first launch:

agy

This opens your default browser for Google OAuth. Sign in with the same Google account you used for Gemini CLI — this matters for plugin and history migration later. For SSH or headless sessions where no browser is available, agy prints a URL directly in the terminal; copy it, authenticate from any device, and the CLI picks up the token automatically once you're signed in.

Step 3: Migrate Plugins, Skills & Config

Gemini CLI extensions migrate with a single command that scans your existing plugin directory and registers each one with agy:

# Import all Gemini CLI plugins/extensions automatically
agy plugin import gemini

# Verify what got imported
agy plugin list

Workspace-local skills need a manual copy — global skills under ~/.config/gemini/skills/ auto-load from the new path, but per-project skills do not:

# Per-workspace skills — copy manually
cp -r .gemini/skills/ .agents/skills/

# Confirm Antigravity picked them up
agy skills list

Good news for one thing

Both GEMINI.md and AGENTS.md are read by Antigravity CLI without any modification — no rename, no reformatting needed. If your team already standardized on AGENTS.md, that convention carries over untouched.

One catch during import: plugins that depend on Node.js-only APIs will trigger a compatibility warning. Those need an updated, Antigravity-compatible build from the original plugin author before they'll work — there's no workaround on your end for those specific ones.

Step 4: Fix Your MCP Server Configs

This is the change most migration guides gloss over, and it's a manual, one-time edit — there's no automated import for it. Gemini CLI stored MCP server configs inline inside settings.json. Antigravity CLI expects them in a separate mcp_config.json file, with the field itself renamed:

Gemini CLI (`settings.json`)	Antigravity CLI (`mcp_config.json`)
Inline under `mcpServers`	Separate file, same key structure
`"url": "https://..."`	`"serverUrl": "https://..."`

If you'd rather not hand-edit the file, you can also re-add each server through the CLI directly:

agy mcp add <name> <command>

Step 5: Fix Scripts, Env Vars & CI/CD

This is the step that actually breaks production if you skip it. Any shell alias, cron job, or CI/CD pipeline step that calls gemini directly needs two separate fixes — not one:

# Before (Gemini CLI)
gemini agents run code-review

# After (Antigravity CLI) — note "agent" not "agents"
agy agent run code-review

Most other commands map one-to-one with just the binary rename (gemini → agy), but grep your codebase for agents separately — that subcommand name itself changed, and a find-and-replace on just the binary name will miss it.

Also update the environment variable your CI/CD secrets reference:

# Old
export GEMINI_API_KEY=your-key-here

# New
export AV_API_KEY=your-key-here

The one thing to prioritize

If you only fix one thing today, fix this. Google's own migration messaging put it plainly: any CI/CD pipeline, shell script, or automation calling gemini "will break on that date with no warning and no grace period." Several developers reported hard, mid-session cutoffs rather than a graceful failure — including one paying user who described the tool actively writing code when it hit a 403 mid-task.

Step 6: Validate the Migration

Before fully cutting over, confirm everything actually works:

# Run a built-in health check
agy doctor

Then run one real workflow through agy that you'd normally run through gemini, side by side if you still have access to the old CLI, to confirm parity before removing Gemini CLI from your CI images or dropping the old binary from your dev environment entirely.

Command Mapping Cheat Sheet

Task	Gemini CLI	Antigravity CLI
Launch the tool	`gemini`	`agy`
Run an agent task	`gemini agents run ...`	`agy agent run ...`
API key env var	`GEMINI_API_KEY`	`AV_API_KEY`
Import extensions	—	`agy plugin import gemini`
Add MCP server	Edit `settings.json`	`agy mcp add <name> <command>`
Session state path	`~/.gemini/brain/`	`~/.gemini/antigravity-cli/brain/`
Health check	—	`agy doctor`

Why This Migration Made Developers Angry

It's worth being upfront about why this rollout generated real backlash on Hacker News and Reddit, beyond the normal friction of any tool migration. Gemini CLI was open source — Apache-2.0 licensed, with over 104,000 GitHub stars and a genuine community of contributors. Antigravity CLI, its replacement, is closed-source.

The core objection: developers who invested time contributing to and improving an open-source tool watched that tool get replaced by a closed product they have no equivalent access to or ownership of. One widely-shared framing from the discussion: Google wasn't accused of building a better enterprise product — it was accused of using community investment in an open-source project to improve a product that then excludes the people who built it.

Practically, this also means: if your team relied on forking or patching Gemini CLI directly for custom behavior, that option doesn't exist with Antigravity CLI. You're now dependent on Google's release cadence and whatever extension/plugin surface they expose.

Should You Migrate at All? The Alternatives

Migrating to Antigravity CLI is the path of least resistance if you're staying inside the Gemini ecosystem — same model family, same account, fastest path back to a working setup. But it's worth knowing what else is on the table, especially given the closed-source concern above:

Claude Code — for terminal-native agentic coding workflows, commonly cited as the practical alternative for individual builders who want to stay in the terminal without adopting Antigravity's IDE-first model.
Aider — for teams that specifically want to keep a fully open-source tool in their pipeline rather than depend on any closed-source vendor CLI.
Google Antigravity (the IDE, not just the CLI) — if your team is willing to go further than a CLI swap, the broader Antigravity platform adds a visual multi-agent Manager Surface, parallel frontend/backend/test/docs agents, and an AGENTS.md-driven "constitution" file for coding standards — genuinely useful if you're coordinating multiple agents on a large codebase, overkill if you just need a terminal assistant.

My honest take, doing this kind of migration work for clients: if you're already deep in the Gemini/Google Cloud ecosystem, Antigravity CLI is the path of least friction and the performance gains (Go vs Node.js, async agents) are real. If you were specifically attached to Gemini CLI's open-source nature, this is a reasonable moment to evaluate Claude Code or Aider instead of assuming you have to follow Google's migration path by default.

A quick note on how I approach tool migrations like this: I build AI automation and AI agent systems for clients, so keeping CI/CD and agent tooling working through vendor changes like this is a regular part of the job — not a one-off inconvenience. If you're dealing with a broken pipeline from this exact migration, or want a second set of eyes on an agent workflow before you commit to a platform, that's exactly the kind of work I do — see my services or check out a few recent projects.

Frequently Asked Questions

Why is my gemini command returning HTTP 410 Gone?

Google shut down Gemini CLI for free, Pro, and Ultra accounts on June 18, 2026. The gemini binary still launches locally, but every API request now returns HTTP 410 Gone because Google stopped serving requests from that endpoint for those account tiers. The fix is to migrate to Antigravity CLI, invoked as agy.

Do I have to migrate to Antigravity CLI?

Only if you're on a free, Google AI Pro, or Google AI Ultra account. Enterprise customers with Gemini Code Assist Standard or Enterprise licenses, and users accessing Gemini CLI through Vertex AI or a paid Gemini API key, retain access and do not need to migrate immediately, though Google is steering all new development toward Antigravity.

Is Antigravity CLI open source like Gemini CLI was?

No. Gemini CLI was open source (Apache-2.0, over 104,000 GitHub stars). Antigravity CLI is a closed-source Go binary. This has been a point of significant community criticism, since Gemini CLI's open-source contributors do not get equivalent access to or ownership of its closed-source replacement.

What is the single most important thing to fix first when migrating?

Any shell script, CI/CD pipeline, or automation that calls the gemini command directly. These fail silently or with a hard 410 error and no advance warning once the cutoff hits. Search your codebase and CI configs for the string "gemini" and replace those calls with agy before touching anything else.

Will my Gemini CLI extensions and MCP servers still work in Antigravity CLI?

Most migrate automatically with agy plugin import gemini, though extensions that depend on Node.js-only APIs will trigger a compatibility warning and may need an updated build from their author. MCP server configs need a manual, one-time edit: move them out of settings.json into a separate mcp_config.json file and rename the url field to serverUrl.