Gemini CLI

Gemini CLI is an open-source AI agent built by Google that runs directly in your terminal, giving developers access to Gemini models for querying and editing large codebases, generating apps from images or PDFs, and automating complex workflows. It is licensed under Apache 2.0 and available via npm, Homebrew, and MacPorts. The project is actively maintained on GitHub and is currently transitioning to a successor product called Antigravity CLI for individual users.

What It Is

Gemini CLI is a terminal-first AI coding assistant that connects to Google's Gemini models (including Gemini 3 with a 1M token context window) from the command line. It is designed for developers who prefer working in the terminal and need direct, lightweight access to a powerful language model without leaving their workflow. The core tool is written in TypeScript and published as an npm package (@google/gemini-cli).

Core Capabilities

The tool ships with a broad set of built-in features:

• Code understanding and generation: Query and edit large codebases, debug issues in natural language, and generate new apps from PDFs, images, or sketches using multimodal input.
• Shell command execution: Run system commands safely from within the agent session.
• Web search and fetch: Ground queries with real-time Google Search results or fetch web content directly.
• File system operations: Read, write, and manage local files and directories.
• MCP (Model Context Protocol) support: Connect to MCP servers to extend the CLI with custom tools and integrations (e.g., GitHub, Slack, databases).
• Agent Skills and Subagents: Use specialized agents for specific tasks, including experimental remote subagents.
• Headless/scripting mode: Run non-interactively in automated pipelines with JSON or stream-JSON output formats.
• Checkpointing and Rewind: Save and resume complex sessions; rewind and replay conversation history.
• Extensions and Hooks: Customize behavior with scripts and community-built extensions.
• Plan mode: A safe, read-only planning mode for complex changes before execution.
• Sandboxing: Isolate tool execution for safer operation.
• IDE integration: VS Code companion support.

Authentication and Access Model

Gemini CLI supports three authentication paths:

• Google Account (OAuth): Sign in with a personal Google account for a free tier (60 requests/min, 1,000 requests/day) with Gemini 3 models and no API key management required.
• Gemini API Key: Use an API key from AI Studio for model selection control and usage-based billing with higher limits.
• Vertex AI: Authenticate via Google Cloud credentials for enterprise security, compliance, SLA, and regional data residency.

Update: Transitioning to Antigravity CLI

Google has announced that Gemini CLI will be replaced by Antigravity CLI for unpaid tier and Google One users. According to a Google Developers Blog post, the transition is driven by users wanting multiple agents working together on a shared backend, prompting Google to consolidate into a single product rather than maintain two separate tools.

Key details of the transition:

• Deadline: June 18, 2026 — Gemini CLI and Code Assist IDE extensions stop working for free, Pro, and Ultra individual users.
• Antigravity CLI is available today and is built in Go (faster), runs agents in the background so the terminal is not locked, and shares the same engine as the Antigravity desktop app.
• Antigravity CLI retains key Gemini CLI features: Skills, Hooks, Subagents, and Extensions (now called "plugins").
• Enterprise customers on paid Standard/Enterprise licenses are not affected — their Gemini CLI access continues as normal.

The full announcement is available at the Google Developers Blog post on transitioning Gemini CLI to Antigravity CLI.

Current Release and Open-Source Status

The latest stable release is v0.42.0 (published May 12, 2026). The project publishes on three channels: stable (weekly, Tuesdays), preview (weekly, Tuesdays, less vetted), and nightly (daily from main). The repository is hosted at google-gemini/gemini-cli on GitHub under the Apache License 2.0, making it fully open source. Community contributions are welcomed via the project's contributing guide and official roadmap.