Short answer
Gemini CLI is Google’s open-source terminal agent, run as gemini. It can work with a codebase and use tools as part of a terminal workflow. If you want an assistant for coding tasks in your terminal, start at the official Gemini CLI repository. Google’s Gemini CLI overview describes local-terminal and Cloud Shell use.
It is not the same as integrating Gemini into your application. For production application code, consult the Gemini API documentation and Google’s GenAI SDK guidance, rather than treating the CLI as your application’s API integration.
Which Google tool fits your task?
- Terminal coding agent: Gemini CLI is intended for terminal-agent workflows, such as understanding or editing a codebase, fixing bugs, or improving tests. Google describes it as using a reason-and-act loop with built-in tools and local or remote MCP servers. These are described capabilities, not guarantees that generated changes will be correct. (Google Cloud overview)
- Stateless API commands from a script, agent, or CI job: The Gemini CLI repository describes a separate Gemini API CLI (
gemini-api) for direct API commands and distinguishes it from a coding harness. Check that tool’s current documentation before adopting it; the repository description is not a full API reference. (Gemini CLI repository) - Production app integration: Use the Gemini API and an official SDK, rather than inferring application API behavior from Gemini CLI features. (Gemini API docs; GenAI SDK guidance)
Suitability warning for newer models
The Gemini CLI repository warns that its harness is not tuned for Gemini 4+ and should not be used to benchmark Gemini models. The repository points readers seeking Gemini 4 Argon+ interactive coding toward Antigravity CLI. Treat this as the repository’s tool guidance, not as independent confirmation of account eligibility or access.
Access changes to check
The repository’s status notice says that, starting June 18, 2026, free individual and Google AI Pro/Ultra access moved to Antigravity CLI. It also describes Gemini CLI as focused on enterprise Gemini Code Assist workflows, while referencing legacy API/Vertex keys elsewhere. Those statements do not fully explain which users or credentials are eligible, so confirm your route in current official documentation before choosing the CLI. (Gemini CLI repository; transition announcement)
That status differs from Google’s June 25, 2025 announcement of a free-access offer. The announcement’s historical limits—60 requests per minute and 1,000 per day—should not be read as current entitlements. (2025 Google announcement)
What can Gemini CLI do—and what does that say about model choice?
Google’s overview names fixing bugs, creating features, and improving test coverage among CLI use cases. The repository also describes querying and editing codebases, debugging, MCP integrations, and scripting. A Google Codelab demonstrates work involving local files, shell commands, web research, and GitHub repositories, with permission prompts for certain actions. These are capability descriptions and tutorial examples, not comparative performance results.
The captured CLI materials show model-selection examples such as gemini -m "gemini-2.5-flash" and the interactive /model command. Treat these as examples, not as a current complete model catalog or evidence that one model is best for coding. For API work, choose a model supported by the current API documentation and evaluate it for your task; the cited CLI materials do not establish a model ranking. (Gemini CLI repository; Google Codelab)
Getting started and choosing an access route
For the documented Cloud Shell route, activate Cloud Shell in the Google Cloud console, navigate to a working directory, then run:
mkdir gemini-cli-projects
cd gemini-cli-projects
gemini
The Google Codelab walks through those steps; Google Cloud says Gemini CLI is available in Cloud Shell without additional CLI setup. Preinstallation does not guarantee that your account is eligible to use the service. For other environments, follow the current installation instructions in the official repository.
Keep the CLI access routes distinct:
- Gemini Code Assist: Google Cloud documentation says edition quotas for Gemini CLI are shared with Gemini Code Assist agent mode. It does not give quota amounts. (Google Cloud overview)
- Gemini API key: Google Cloud separately describes API-key use as pay-as-you-go. That is not the same as a Code Assist quota; the overview does not provide current rates or billing units. (Google Cloud overview)
- Vertex AI: The CLI repository shows Vertex AI as an authentication option, but the materials here do not establish a current price comparison or a complete production setup for an application. (Gemini CLI repository)
For application code, Google’s text-generation guide shows this JavaScript SDK pattern. This is a Gemini API example—not a Gemini CLI command:
import { GoogleGenAI } from "@google/genai";
const ai = new GoogleGenAI({});
const result = await ai.interactions.create({
model: "gemini-3.8-flash",
input: "How does this service handle authentication?",
});
console.log(result.output_text);
See Google’s API-key guidance for credentials. The cited CLI and overview pages do not establish current pricing, billing units, or exact quotas, so they are not enough to make a cost comparison.
Bottom line: use Gemini CLI for an eligible terminal-agent workflow, a dedicated API CLI for stateless API commands, and the Gemini API with an official SDK for production application integration. Check current account eligibility and API terms before settling on an access or cost plan.
Sources
Gemini CLI official repository
Gemini CLI | Gemini for Google Cloud
Important Update: Gemini CLI Has Stopped Serving Requests for Individual Accounts