# Antigravity CLI > Run Google's Antigravity CLI (agy), the successor to Gemini CLI, in NeuroSquad cards — how to install it, what works, and what doesn't. Source: https://docs.neurosquad.ai/en/agents/antigravity **Antigravity CLI** (`agy`) is Google's terminal coding agent and the successor to Gemini CLI: one program that shares its agent engine with the Antigravity desktop app. It is closed source — its GitHub repository holds the README and the changelog, not the code. In NeuroSquad each Antigravity agent is a card with the real `agy` running inside. Measured on Antigravity CLI 1.3.1. ## Install **1. Install the agy command** Use Google's installer. On Windows, in PowerShell: ```powershell irm https://antigravity.google/cli/install.ps1 | iex ``` On macOS and Linux: ```bash curl -fsSL https://antigravity.google/cli/install.sh | bash ``` **Settings → Setup** in NeuroSquad can install it for you. **2. Sign in once** Run `agy` in a terminal and sign in with your Google account — or set it up with a Gemini API key (`modelProvider: "gemini"` in its settings and `GEMINI_API_KEY` in your environment). Its first screens — the colour scheme and Google's terms and data-use consent — are yours to answer; NeuroSquad never answers them for you. A card can also run on [OpenRouter](https://docs.neurosquad.ai/en/providers/openrouter) or on [your own server](https://docs.neurosquad.ai/en/providers/your-servers) instead. **3. Add an Antigravity card** Add an agent card and choose **Antigravity CLI**. ## What works - **Exact status.** agy's own lifecycle hooks tell the card when a turn starts (`PreInvocation`) and ends (`Stop`); the state agy hands its window-title command says when a confirmation is on screen, and the card names what it wants to run (`run_command — rm -rf dist`) or the question it asks. See [Finished / needs your input](https://docs.neurosquad.ai/en/agents/notifications). - **Arrows.** NeuroSquad's MCP server is in the card's own `mcp_config.json`, allowed by a `mcp(neurosquad/*)` rule, so its tools never prompt. agy re-lists the tools when they change: an arrow drawn while it works is in its very next request. Installed MCP servers come through separately, and agy asks before every call to them. - **Canvas mode.** A NeuroSquad hook denies `run_command`, `manage_task`, `search_web`, `read_url_content` and the browser tools, so commands run in terminal cards — in dangerous mode too. See [Canvas mode](https://docs.neurosquad.ai/en/squads/canvas-mode). - **Sessions.** agy picks the conversation id; the card learns it from the hooks and continues the same conversation after a restart (`agy --conversation `). `/clear` moves the card to the new conversation. A conversation that can't be found starts fresh by itself. - **Long sessions.** Context meter (against the window agy reports), the prompt queue, the journal and hand-off. agy compacts its context by itself. - **Usage.** Every request's tokens — input, cached, output — are read from agy's own transcript; subagents are conversations of their own, counted with their parent and marked as such, in [Usage & costs](https://docs.neurosquad.ai/en/usage), [Run Stats](https://docs.neurosquad.ai/en/cards/run-stats) and [Agent Pulse](https://docs.neurosquad.ai/en/cards/agent-pulse). - **Models and providers.** Pick the model per card; a lead agent can change it with `agent_set_model`. OpenRouter and your own OpenAI-compatible servers work through a small translator NeuroSquad runs on your machine (agy speaks only the Gemini API); the key never reaches agy. - **Switching CLIs.** Move the conversation to Claude Code and back. - **Plugins.** [Memory](https://docs.neurosquad.ai/en/plugins/mem0-memory), [Context7](https://docs.neurosquad.ai/en/plugins/context7-docs), [Caveman](https://docs.neurosquad.ai/en/plugins/caveman) and [Graphify](https://docs.neurosquad.ai/en/plugins/graphify) reach agy's turns through its `PreInvocation` hook. - **From your phone.** The card works in [remote access](https://docs.neurosquad.ai/en/remote) like any other. ## Dangerous mode A card in dangerous mode starts agy with `--dangerously-skip-permissions`. agy has no way to switch it while it runs, so turning it on or off restarts the card on the same conversation. See [Dangerous mode](https://docs.neurosquad.ai/en/agents/dangerous-mode). ## Coming from Gemini CLI Google closed Gemini CLI to free and Pro/Ultra users on 2026-06-18, and Antigravity CLI took its place in NeuroSquad's add-card menu. Gemini cards you already have keep running, marked **Gemini CLI (legacy)**; **Switch to Antigravity** on the card moves the conversation over in one click. Templates that name Gemini CLI open as Antigravity CLI. agy reads your `GEMINI.md` and `AGENTS.md` as Gemini CLI did. ## Good to know - **Your `~/.gemini` is never written.** Each card runs agy with a Gemini folder of its own: its settings, `hooks.json` and `mcp_config.json` are your own plus NeuroSquad's layer, and your conversations, onboarding, plugins, skills and rules are linked in. So `agy --conversation ` works from a plain terminal too. Settings you change inside a card stay with the cards. - **Stop a turn with Esc.** Ctrl+C on an idle agy only arms its exit, so NeuroSquad never sends it — the budget brake and the phone's key row use Escape. - **agy's self-update is off inside cards.** Update it from a terminal (`agy update`). - **No price for Google's own models.** NeuroSquad has no Gemini list prices yet, so cards on your Google sign-in show tokens and "no price" — never $0. agy's extra call that titles a conversation isn't in its transcript, so Usage doesn't count it. - **No token saver.** The [token saver](https://docs.neurosquad.ai/en/plugins/rtk-ai-token-saver) rewrites shell commands; agy's hook can allow, ask or deny a command but not change it. - **Your data and Google.** agy runs under your Google account and Google's terms; Google collects Interactions data unless you opt out in agy's settings — see [antigravity.google/terms](https://antigravity.google/terms). > Antigravity cards run on this computer only — not yet in WSL or SSH workspaces.