Codex CLI
Codex CLI is an open-source command-line coding assistant tool released by OpenAI, built on large language models and designed to run in terminal environments. It interprets natural language instructions to automatically generate, modify, and explain code, and can execute operations on the local file system to assist developers with coding, debugging, and refactoring tasks. It supports multiple operating modes and allows users to configure different model backends to suit various development scenarios.
Core Facts
MCP采用客户端-服务器架构，基于标准化的JSON-RPC通信协议
Codex CLI can read local codebases, execute shell commands, edit files, and run tests.
Codex CLI is a command-line programming tool launched by OpenAI that allows developers to interact directly with AI models in the terminal to complete coding tasks.
Timeline (last 90 days)
The experiments evaluated four general-purpose harnesses (OpenCode, Codex CLI, Pi CLI and DeepAgents) paired with four core models, plus two fixed native model-harness systems, covering six benchmarks.
At Dev Day on September 29, 2026, OpenAI announced Codex Cloud, which moves Codex sessions into OpenAI-hosted containers so tasks continue running after the laptop is closed and can be viewed from a phone; Codex CLI also added a /agents view and a /fork command.
Google Cloud announced a private preview of three models for select enterprise customers, who can call them via API and also connect to the Gemini platform for inference through local Codex CLI or Desktop workflows.
/presets provides ten nameable configuration slots for saving a set of model and permission parameters and quickly invoking them via the slots.
Before 0.164, /model and /permissions were mostly invoked without arguments, and after execution entered a submenu or a follow-up style selection.
Version 0.164 allows parameters to be attached directly when calling /model and /permissions, eliminating the second interaction.
Version 0.164 added a credential masking mechanism in remote command scenarios, masking sensitive information when command output or logs may contain it.
Codex CLI's design typically includes a diff preview step that lets users confirm the AI-generated changes before committing them.
Codex CLI's shortcomings include being unfriendly to beginners, and the quality of changes made by the local agent depends heavily on model capability and project complexity.
Codex CLI recommends keeping the workspace clean before starting a task so that all changes can be reviewed at once via git diff and, if necessary, discarded entirely with git checkout.
40 more timeline events
All Facts (20)
MCP采用客户端-服务器架构，基于标准化的JSON-RPC通信协议
90%VerifiedCodex CLI can read local codebases, execute shell commands, edit files, and run tests.
80%VerifiedCodex CLI is a command-line programming tool launched by OpenAI that allows developers to interact directly with AI models in the terminal to complete coding tasks.
80%VerifiedCodex CLI offers three permission levels: suggest mode, auto-edit mode, and full-auto mode.
80%VerifiedCodex CLI uses the TOML format as its configuration file.
75%VerifiedCodex CLI is distributed as an npm package and requires a Node.js runtime environment.
70%VerifiedGemini CLI belongs to the same category of Agentic Coding tools as Claude Code and OpenAI Codex.
70%VerifiedCodex CLI was officially opened in 2025, allowing developers to invoke it directly in a local terminal.
70%VerifiedGit's post-checkout hook triggers after git checkout, git clone, and git worktree add; when a new worktree is created via git worktree add, Git passes an all-zero null ref as the first argument.
65%VerifiedThe command-line interface (CLI) is the most fundamental human-computer interaction method in operating systems, where users control program execution by typing text commands.
65%VerifiedCodex CLI is slower when performing code reviews and is suitable for handling complex changes or backend tech stacks.
65%VerifiedThe Codex CLI 0.154.0 update adds Windows daemon support.
65%UnverifiedCode Review Router的Codex CLI硬性规则包括：文件数>20、行数>500、涉及数据库迁移/API服务层修改、跨3个顶级目录等
95%UnverifiedCode Review Router Skill的路由决策中，复杂度得分≥6分由Codex深度分析，<6分由Gemini快速反馈
95%UnverifiedCodex CLI项目在GitHub上开源，社区贡献者可以直接参与功能开发
90%Unverified/goal 的自主循环中，随着循环轮次增加，单轮的 Token 消耗会加速增长（除非实施上下文压缩策略）
90%UnverifiedCodex CLI通过token预算机制控制自动循环的资源消耗，用户可配置token使用上限
90%UnverifiedCodex CLI支持Full Auto（全自动）和手动审批两种工作模式
90%UnverifiedCodex CLI通过API与OpenAI云端模型通信，采用云端大脑加本地执行的架构
90%UnverifiedCodex CLI的配置核心是两个文件：Config TOML负责模型和接口地址配置，Auth文件负责API Key认证信息
90%