AI library

Source · docs

Gemini API CLI (gemini-api) beta

Author
Google
Published
27 October 2026
Added
28 October 2026
Original
https://aistudio.google.com/docs/cli

Summary

On 2026-10-27 Google released gemini-api, an open-source beta command-line client for the Gemini API. It is a single binary with no dependencies and no state between runs, built on the Interactions API and aimed at coding agents, scripts, and CI. It can generate text, images, speech, music, and video, analyse files, and run Gemini managed agents such as Deep Research from the shell. It is a different tool from Gemini CLI, Google's coding agent.

Key points

  • Install: curl -fsSL https://raw.githubusercontent.com/google-gemini/gemini-api-cli/main/scripts/install.sh | bash on macOS or Linux, or go install github.com/google-gemini/gemini-api-cli/cmd/gemini-api@latest. Pin a version with GEMINI_VERSION in the install script.
  • Authenticate: export GEMINI_API_KEY=... or gemini-api configure --api-key ... (stores it in the OS keychain). gemini-api whoami shows which credential is in use.
  • Commands: generate, image, tts, music, video, analyze, transcribe, agent run|status|cancel, plus files, models, environments, webhooks, and triggers. Media commands write files and print the path.
  • Agent-friendly output: -o json, -o toon, or --jq for machine-readable output, --dry-run to preview a request, and --schema to inspect payloads before building --body. Streaming commands need --stream=false with JSON output.
  • Using it from Claude Code or Codex: there is no skill package; the guide suggests adding a short snippet to CLAUDE.md or AGENTS.md telling the agent to read gemini-api --help, check --schema, and use --dry-run -o json for complex calls.
  • Example: gemini-api image "isometric cloud database icon" --out ./db.jpg, or gemini-api agent run "Compare WebRTC SFUs" --agent deep-research-preview-04-2026 --background --stream=false.
  • Limits: inline inputs up to 14 MB raw (use files upload --wait for larger), uploaded files are deleted after 48 hours, no WebSocket or Live API support. Exit codes: 0 ok, 1 runtime or API error, 2 usage error, 3 auth error.
  • Beta: breaking changes can happen between minor versions, so pin a release in CI.

Why it matters

It gives any coding agent shell access to Gemini's image, video, speech, and Deep Research features without an MCP server or SDK code. A Claude Code session can, for example, generate an asset or start a Deep Research job with one Bash command.