Google Unveils Antigravity CLI: A Go-Powered, Free AI Agent Successor to Gemini CLI

Google has officially released Antigravity CLI, a brand-new command-line interface (CLI) agent touted as the continuation of the Gemini CLI. While initially appearing as a mere rebranding, Antigravity CLI represents a complete rewrite in Go, delivering significant performance improvements, including notably fast response times and command execution. The agent is available for free with a Google plan, allowing users to leverage models like Gemini Flash 3.5 (both high and low versions), Opus, Cloud Sonnet, and even an open-source GPT variant.

Antigravity CLI offers a comprehensive suite of features designed to streamline development. Installation is straightforward across Windows, macOS, and Linux, with authentication via Google account or Google Cloud API tokens. Its core capabilities include rapid code generation (demonstrated with Kanban applications in HTML and React, including theme implementation), a /artifacts command for reviewing project modifications, and the /btw (by the way) feature for concurrent questioning without interrupting ongoing tasks. Developers can configure tool permissions, utilize a /planning mode for structured task execution (e.g., implementing backends with Hono and SQLite), and easily switch between AI models via the /model command. Advanced features extend to /usage for tracking token consumption, /context for visualizing the context window, /rewind for session rollbacks, and /clear to reset the context. The agent also supports direct bash command execution (#) and multi-line input with Ctrl+Enter.

Further enhancing its utility, Antigravity CLI incorporates an /agents feature, allowing the creation of specialized sub-agents (e.g., ‘research’ and ‘tester’ agents) for parallel and focused task execution. It supports the integration of external AI skills, such as Anthropic’s ‘Frontend Design,’ to improve UI/UX generation, and includes a /goal command for iterative task completion until a defined objective is met. Despite its current lack of official documentation, initial impressions highlight its fluid operation and competitive performance against many paid AI agents, even on its free tier. Future updates are expected to include advanced features like MCPs, a /schedule command for task programming, and hooks for task-completion events, alongside notification support.