CommandInstalls cleanly

Make images and transcripts with Gemini

Gemini API CLI

Google's gemini-api command lets Claude Code, Codex or any shell agent make images, voiceovers and transcripts, read PDFs and run Gemini research agents.

Works with
Claude Code, Codex CLI, Any agent
Good for
Graphic designers, Daily work and research

What it does

gemini-api is one small program from Google that gives any assistant with a terminal (Claude Code, Codex, Antigravity and others) access to Gemini models. Your assistant plans the work and calls gemini-api when it needs something Gemini is good at:

  • image: make or edit images and save them in your project.
  • tts and music: make voiceovers and background music.
  • video: make or edit video clips.
  • analyze: answer questions about PDFs, images, audio, video, CSV or text files, or public YouTube links.
  • transcribe: turn a recording into Markdown, plain text, JSON or SRT captions.
  • agent run: start a Gemini managed agent, such as Deep Research, in the background and fetch the result later.

It is not a chat app and not a coding assistant. Each command runs once and prints a file path or an answer, which the assistant then uses. Google calls it beta.

If you do not want to pipe a script into bash, download the archive for your computer from the GitHub Releases page, compare it with the release's checksums.txt, unpack it and put gemini-api in a folder on your PATH. Google's docs also show how to install one exact version: add GEMINI_VERSION=<release-tag> before bash in the install line.

Example prompts

  • "Use gemini-api to make a friendly robot mascot for this landing page and save it in ./assets/."
  • "Transcribe interview.mp4 with gemini-api into SRT captions."
  • "Use gemini-api analyze on product-brief.pdf and tell me what we should build first."
  • "Start a Gemini deep research run on battery recycling with gemini-api and check on it in ten minutes."

How to install

curl -fsSL https://raw.githubusercontent.com/google-gemini/gemini-api-cli/main/scripts/install.sh | bash

How to use it

Set GEMINI_API_KEY, add the README's short instruction block to CLAUDE.md or AGENTS.md, then ask: Use gemini-api to make a mascot image for this site and save it in ./assets/.

Check it works

In a terminal, run gemini-api generate "Hello from my agent" --dry-run. It checks your setup and shows the request without a network call. Then run it again without --dry-run to call the API. gemini-api whoami shows your active settings and where your key comes from.

Common problems

Other assistants

  • Codex and other agents: the same program works for any agent with shell access; add the README's instruction block to AGENTS.md (README).
  • Windows (PowerShell): iwr -useb https://raw.githubusercontent.com/google-gemini/gemini-api-cli/main/scripts/install.ps1 | iex (README). This script also does not check checksums.
  • With Go: go install github.com/google-gemini/gemini-api-cli/cmd/gemini-api@latest (README).
  • Gemini CLI: no primary source gives a Gemini CLI setup line. The Gemini CLI README's "Which Google CLI do you need?" table lists gemini-api for agents such as agy, Claude Code and Codex, scripts and CI, and lists Gemini CLI itself for organizations with Gemini Code Assist Standard or Enterprise.

What to watch out for

  • Beta: the README says breaking changes may occur between releases and recommends pinning to a specific release version in CI and automated workflows (README, Status section)
  • Every real call uses your Gemini API key. New keys start on the Free Tier, which gives access to certain models up to their free rate limits; some models are not available on the Free Tier and paid use is billed per token (ai.google.dev billing and pricing pages)
  • On the Free Tier, Google's pricing page marks content as used to improve Google's products; on the Paid Tier it is not (ai.google.dev/gemini-api/docs/pricing)
  • Files and prompts you send go to Google's API. Uploaded files are deleted after 48 hours. Interactions are stored by default, for 55 days on the Paid Tier and 1 day on the Free Tier, unless you send "store": false in a --body request (README, Files API and Interactions API docs)
  • The install line downloads a script from GitHub and runs it with bash. The script (from the main branch) installs the latest release and does not check the checksums.txt file that each release ships; read scripts/install.sh first, or download a release archive and check it yourself (scripts/install.sh, GitHub Releases)
  • The README gives you a text block to paste into CLAUDE.md or AGENTS.md. It tells the agent to inspect --help or --usage first and to 'Preview API calls with --dry-run'; read it before you add it (README, Quick start step 2)

Licence and author

Licence
Apache-2.0
Author
Google
Source code
github.com/google-gemini/gemini-api-cli

How we tested it

We installed it on an empty test computer on 28 Oct 2026. Installed and started in a clean test system. Not yet used in a real AI session.

Test level

No API key, so no calls to Google were made. Not yet used on real work.

Technical details of the test
Date
28 Oct 2026
Method
Manual sandbox run on 2026-10-28 in the intake's clean container with no account: read scripts/install.sh, ran it as a normal user (it installed v0.4.4 to ~/.local/bin because /usr/local/bin was not writable), then ran "gemini-api --version" and "gemini-api --help".

Checked again on 28 Oct 2026

Alternatives

  • Nano Banana — image commands inside Gemini CLI instead of a separate program
  • HyperFrames — make videos from HTML instead of generating them with a model
  • imagegen-skills — skills that guide image generation for mockups and brand boards

FAQ

  • Q: Does it cost money? A: The CLI is free (Apache-2.0). Calls use your Gemini API key. New accounts start on Google's Free Tier, which covers certain models up to their rate limits; other use is billed under Google's Gemini API pricing.
  • Q: Is this the same as Gemini CLI? A: No. Google's docs say gemini-api "is a stateless API client, not an interactive coding agent like @google/gemini-cli". It is a tool your existing assistant calls.
  • Q: Can I try it without spending anything? A: Yes. --dry-run checks the request without calling the API or reading your keychain (README). After that, a Free Tier key lets you make real calls to some models at no charge, but on the Free Tier Google may use your content to improve its products.

Sources