AI coding agents that run in the terminal have become everyday tools for developers and sysadmins. The three big AI companies each offer one: Anthropic’s Claude Code, OpenAI’s Codex CLI and Google’s Antigravity CLI, which replaced Gemini CLI for individual users in June 2026. All three can read a codebase, edit files and run commands with your approval. The differences are in how you install them, how you pay, and how they fit into your workflow.
This comparison focuses on what matters on Linux: installation, sign-in options, project instruction files, automation and extensibility. Model quality changes with every release, so test each agent on your own tasks before you commit.
Quick Comparison
| Claude Code | Codex CLI | Antigravity CLI | |
|---|---|---|---|
| Company | Anthropic | OpenAI | |
| Command | claude | codex | agy |
| Linux install | Native installer script, or apt/dnf/apk repositories | npm (Node.js 22+) or standalone installer script | Installer script |
| Needs Node.js | No | Only for the npm install | No |
| Sign-in | Claude Pro/Max/Team/Enterprise, or Console API billing | ChatGPT plan or OpenAI API key | Google account |
| Open source | No | Yes | See the project repository |
| Project instruction file | CLAUDE.md | AGENTS.md | See the documentation |
| Non-interactive mode | claude -p | codex exec | See agy --help |
| MCP servers | Yes (claude mcp add) | Yes (config.toml) | See the documentation |
| Install guide | Claude Code on Linux | Codex CLI on Linux | Antigravity CLI on Linux |
Installation and Updates
All three install for your user account without root. Claude Code and Antigravity CLI ship as single binaries from an install script and update themselves; Claude Code also offers signed package repositories, which suits servers managed with apt or dnf. Codex CLI is the only one available from npm, which is convenient if Node.js is already part of your toolchain, and it also has a standalone installer.
Pricing and Sign-In
Each agent can use the subscription you may already pay for: a Claude plan, a ChatGPT plan, or a Google account. Claude Code and Codex CLI can also bill per token through an API key, which is better for automation and shared servers because usage is tracked separately from personal chat use. Plans and limits change often, so check each vendor’s pricing page before choosing.
Working Style and Safety
All three ask for approval before they run commands or change files, and all three work best inside a Git repository, where you can review changes with git diff and undo them. Codex CLI runs commands inside an operating-system sandbox by default. Claude Code uses a permission system that you can tune per project in .claude/settings.json, for example to always allow npm test but never rm.
Whichever you choose, never run an AI agent as root on a production server, and keep secrets out of the folders it can read.
Automation and Extensibility
Claude Code (claude -p) and Codex CLI (codex exec) both have a non-interactive mode that you can call from shell scripts, cron jobs or CI pipelines, for example to summarize logs or review a pull request. Both connect to Model Context Protocol (MCP) servers, which give the agent access to tools such as GitHub, a browser or your monitoring system; see the best MCP servers for developers and sysadmins.
Which One Should You Choose?
- Choose Claude Code if you already have a Claude plan, want package-manager installs on servers, or want fine-grained permission rules per project.
- Choose Codex CLI if you already pay for ChatGPT, want an open-source agent you can inspect, or like its default sandbox.
- Choose Antigravity CLI if you work in Google’s ecosystem or used Gemini CLI before and want Google’s supported successor.
- Want everything local? These agents use cloud models. To keep code on your own hardware, run a local model with Ollama or vLLM and use a tool that supports local models.
Many developers install two of them and use whichever handles a given task better. Because all three read plain-text instruction files, keeping both a CLAUDE.md and an AGENTS.md in a repository costs little.
How did it make you feel?
