Streaming output (token-by-token):

gemini --stream "Write a limerick about cats"

npm-gemini

Run Google Gemini from your terminal. Fast CLI with chat mode, Markdown output, and simple config.

Install

Global install (recommended):

# from the project root
npm install -g .

Or use it locally with npx while developing:

npx --yes . --help

Authenticate

You need a Gemini API key. Get one from:

• https://aistudio.google.com/app/apikey

Then set it once:

gemini auth YOUR_API_KEY

Alternatively, set an environment variable (takes priority):

export GEMINI_API_KEY=YOUR_API_KEY

Usage

Basic prompt:

gemini "Explain webhooks like I'm five"

Pipe from stdin:

# prompt can be provided as args or omitted when piping
cat README.md | gemini "Summarize this file"
cat README.md | gemini

Choose a model and temperature:

gemini -m gemini-1.5-pro -t 0.4 "Draft a professional email about a shipping delay"

Raw JSON output:

gemini --json "Return a JSON array of 5 startup ideas with fields: name, one_liner"

Disable Markdown rendering:

gemini --no-md "Show raw text only"

Interactive chat:

gemini chat

With a system instruction:

gemini chat --system "You are a concise, senior TypeScript engineer."

Config

View current config:

gemini whoami

Get config JSON:

gemini config --get

Set defaults:

gemini config --set model=gemini-1.5-pro temp=0.4 system="Be terse"

Config is stored at:

• ~/.config/gemini-cli/config.json

Environment variable GEMINI_API_KEY overrides the stored key.

Programmatic API

You can also import the library methods if you install this package as a dep:

import { generateText } from 'npm-gemini';

const { text } = await generateText({
  apiKey: process.env.GEMINI_API_KEY,
  model: 'gemini-1.5-flash',
  temperature: 0.7,
  prompt: 'Write a haiku about JavaScript',
});
console.log(text);

Notes

• Requires Node.js 18+.
• Uses the official @google/generative-ai SDK.
• Markdown is rendered in the terminal via marked + marked-terminal. Use --no-md to print plain text.
• For issues or feature requests (e.g., response streaming), open an issue.