Gemini API CLI

Gemini API CLI (gemini-api) is a command-line tool from the google-gemini organization that connects agents with shell access, such as Codex, Claude Code, Antigravity, OpenClaw, and Hermes, to the Gemini API. The README describes it as beta software, and the latest listed release is v0.4.3. It is distributed as a single Go binary for macOS, Linux, and Windows.

What It Is

The CLI wraps the Gemini API in standard shell commands. A host agent or script configures an API key, then calls commands to generate text, images, speech, music, and video, analyze files, transcribe recordings, or launch managed agents. Each invocation is stateless: the calling agent supplies context, chooses the model, and decides the next step. Media commands write files to disk and print the resulting path, while agent runs return interaction IDs that can be polled later.

Agent-friendly interface

The binary describes itself so agents can inspect it without external documentation. --usage emits the command tree as KDL, --schema prints JSON Schema for request bodies, and --dry-run previews the resolved HTTP request with credentials redacted and no network call. Output can be streamed text, JSON, TOON, or filtered inline with --jq. An agent mode produces structured JSON error envelopes, and distinct exit codes separate runtime, usage, and authentication failures.

Workflows covered

Commands span creation (generate, image, tts, music, video), understanding (analyze for PDFs, images, audio, video, CSV/text, and YouTube URLs; transcribe to Markdown, text, JSON, or SRT), and management of agents, files, models, environments, credentials, webhooks, and cron triggers. Long-running work can be delegated to a managed agent such as Deep Research in the background and checked later by interaction ID. For full control, --body accepts the underlying Interactions API request schema.

Current Status

The README states the CLI is in beta and evolving alongside the Gemini Interactions API, so breaking changes may occur between releases and pinning a release version is recommended. The project is Apache 2.0 licensed and is not accepting pull requests at this time.