Google has officially unified its terminal-native developer ecosystem. Following announcements at Google I/O 2026, Google has retired the legacy Node-based Gemini CLI and replaced it with the Antigravity CLI, a high-performance command-line coding agent built from the ground up in Go.
Designed specifically for remote server environments, SSH-heavy workflows, and developers who prefer to stay entirely within the terminal, the new interface brings multi-file editing, asynchronous subagents, and Model Context Protocol (MCP) integrations directly to the terminal prompt. By executing via a binary written in Go, the application boasts a near-instant startup time and a fraction of the memory overhead seen in its predecessor.
What is the Antigravity CLI?
The Antigravity CLI (agy) functions as a dedicated, terminal-based alternative to the newly released Antigravity 2.0 desktop development environment. Rather than acting as a superficial chat window wrapper around an LLM API, the CLI shares a unified “agent harness” engine with the desktop application.
This means that any updates to the core multi-step reasoning capabilities, tool-calling systems, or system permissions instantly apply to both platforms. Developers can even begin a complex, multi-file troubleshooting session inside the terminal and use the /export command to seamlessly push the active session over to the desktop GUI without losing context.
Key Architectural Upgrades: Asynchronous Subagents
The most significant workflow upgrade within the Antigravity CLI is its built-in Asynchronous Subagents framework. When handed a sprawling, multi-step goal (like restructuring an entire repository’s testing suite), the primary agent doesn’t lock up your terminal.
Instead, the parent agent spins up multiple background subagents tasked with isolated, parallel components of the workload:
┌──────────────────────┐
│ Parent Agent (TUI) │ ──► Keeps terminal input active
└──────────┬───────────┘
│
Spawns parallel subagents
│
┌───────────────────┴───────────────────┐
│ │
▼ ▼
┌─────────────────┐ ┌─────────────────┐
│ Subagent A: │ │ Subagent B: │
│ Refactoring Lib │ │ Writing Unit │
│ Workspace │ │ Tests │
└─────────────────┘ └─────────────────┘
Developers can type /agents at any time to open an interactive monitoring panel, check live tool execution logs, or use the Ctrl + J shortcut to “teleport” directly into a subagent’s detailed view. If a background agent needs permission to run a local command or modify a protected file, it surfaces a Fast Path Alert directly above your active prompt block, allowing for instant Ctrl + K approval without forcing you to switch tabs.
Technical Comparison: The Terminal Agent Landscape
Google’s rollout places it in a head-to-head competition with Anthropic’s Claude Code and OpenAI’s Codex CLI for terminal dominance. The core technical metrics highlight where the Go-based rewrite shifts the balance:
| Technical Metric | Google Antigravity CLI (agy) | Anthropic Claude Code | Legacy Gemini CLI (Retired) |
| Underlying Language | Go (Compiled Binary) | TypeScript (Node.js) | TypeScript (Node.js) |
| Default Model Engine | Gemini 3.5 Flash / Pro | Claude 3.5 Sonnet / 4.6 Opus | Gemini 1.5 Pro |
| Authentication Flow | SSH-Aware Keyring + Remote URL | Local OAuth Web Browser | Standard API Key Variable |
| Security Layer | Native Sandbox (nsjail/sandbox-exec) | User-defined permissions | None (Runs bare on host) |
Enterprise Security: The Terminal Sandbox
Allowing an autonomous agent to execute shell commands directly on a production server or local machine poses immense security risks. To mitigate this, Google has built a zero-overhead Terminal Sandbox directly into the CLI.
Instead of spinning up slow virtual machines or system containers, the agy binary interfaces with native operating system containment protocols: nsjail on Linux, sandbox-exec on macOS, and AppContainer on Windows. This places strict write and outbound network constraints around the agent’s workspace. If a developer needs to run a trusted command outside the sandbox, the tool surfaces an interactive prompt allowing a manual bypass for that specific action.
How to Install and Initialize the Antigravity CLI
Setting up the terminal interface requires running a single installation command depending on your target system environment.
1.Run the Automated Setup Script:Under 30 seconds.
Open your terminal wrapper and execute the official installation pipeline:
- macOS & Linux:
Bash
curl -fsSL https://antigravity.google/cli/install.sh | bash
- Windows (PowerShell):
PowerShell
irm https://antigravity.google/cli/install.ps1 | iex
2.Execute the Unified Login Sequence:OAuth Keyring Cache.
Initialize the security handshake by running the root binary command:
Bash
agy
On a local machine, this will automatically launch your default browser to complete a secure Google Sign-In.
(Note: If you are connected over a remote SSH pipe, the CLI will intelligently detect the headless environment, print a short authorization URL, and supply a one-time activation code to paste into your local browser).
3.Configure Your Workspace Context:Optional Settings File.
Navigate into your active repository. You can instantly configure preferred model fallbacks, register custom external tool shortcuts, or connect enterprise Model Context Protocol (MCP) servers by adding your settings definitions directly inside:
Bash
~/.gemini/antigravity-cli/settings.json
Pro Tip: Drop a plain-English instructions file titled
AGENTS.mdat the root directory of your project repository. The Antigravity CLI parses this file automatically upon initialization, letting you lock down custom architectural rules, banned file tracks, or styling preferences that subagents must follow during automated refactors.
