Google released an experimental preview of the Gemini API CLI, a zero-dependency binary named gemini-api, for developers and for coding agents that already have a shell. From the terminal it can generate text, images, video, speech, and music, analyze documents and media, transcribe audio, and start managed agents such as Deep Research. The same commands work inside Claude Code, Codex, Cursor, Antigravity, and other agents, and in ordinary scripts and CI jobs. Agents that needed those capabilities often had to guess model IDs, write throwaway scripts, or pull raw API responses and base64 into their context. gemini-api stays stateless. Media commands write the file to disk and print only the path. The binary lists its own commands and schemas offline, and a dry-run option previews the request with credentials removed before any network call. It runs on macOS, Linux, and Windows and uses a Gemini API key from Google AI Studio. Current defaults include Gemini 3.8 Flash for text and analysis, Gemini Nano Banana 2.1 for images, Gemini Omni 1.1 Flash for video, Lyria 3.5 for music, and Gemini 3.8 Flash TTS for speech. The CLI is a separate layer from the Google Gen AI SDKs used in production Python, TypeScript, Go, and Java services, and from the interactive Gemini CLI coding agent. This v0.4.x preview is open source under Apache 2.0 and is still changing alongside the Interactions API. Google is collecting feedback on the command surface in the project repository.