Gemini
Gemini. Is Google’s open-source agentic coding agent for the terminal, ReAct loop, built-in tools (files, shell, search), MCP servers, and large-context Gemini models.
Summary
Gemini CLI (google-gemini/gemini-cli) runs an agent loop in your repo: reason, call tools, checkpoint conversations, and load project context via GEMINI.md. It shares quota with Gemini Code Assist agent mode on free and paid Google tiers.
Use for interactive coding, refactors, and skill-driven workflows when Gemini models or Google Cloud billing are the constraint. It is a repo-bounded AI Agent; not an omnichannel personal bot (OpenClaw, Hermes are hold). For scheduled automation, prefer deterministic scripts and Dagu with LLM steps only where reasoning is required.
Compare Claude Code (Anthropic) and Codex (OpenAI) for the same agent-loop-in-repo pattern.
Details
- Install:
npm install -g @google/gemini-cli, Homebrew (brew install gemini-cli), ornpxfrom GitHub; Node.js 20+. - Auth: Google account (free tier quotas) or Gemini API key for pay-as-you-go.
- Models: Gemini 2.5 Pro and successors; very large context windows per Google docs.
- Skills: SKILL.md / Agent Skills Framework ecosystem, see Agent Skills - Sources for cross-agent collections.
- Fit: Tool / AI Agent, agent-loop coding tool (also branded Gemini Code Assist in IDE surfaces).
- Contrast: Cursor + cursor-agent for default IDE workflow; Claude Code / Codex on other vendors.