Code, a community-driven fork of OpenAI/codex (github.com)

Code is a community-driven fork of openai/codex that turns the CLI codex experience into a faster, developer-centric local coding agent. The release adds browser integration (Chrome CDP and headless browsing with screenshots), a side-by-side diff viewer, a theme system, and multi-agent orchestration commands (/plan, /solve, /code) that can run agents in consensus or race modes. It preserves upstream compatibility while adding reasoning-control (/reasoning), safety modes (read-only, approval workflows, sandboxing), and MCP extensibility so you can plug in filesystem, DB, API or custom tool servers. Technically, Code supports ChatGPT sign-in or API-key usage (creds stored locally at ~/.code/auth.json), works with multiple model providers (default example model = "gpt-5"), and can orchestrate other AI CLIs (Anthropic Claude, Google Gemini, Qwen) after installing companion CLIs. Configuration lives in ~/.code/config.toml (backwards-read of ~/.codex/ for compatibility), and MCP servers are declared as named tables for tooling like modelcontextprotocol filesystem servers. The CLI is distributed via npx/npm, requires Node.js 20+, and is Apache-2.0 licensed (not affiliated with OpenAI). For engineers and ML practitioners this means a portable, auditable agent for local dev, CI automation, and multi-model experiments with configurable safety and reasoning effort—useful for reproducible workflows, tool-augmented code generation, and research into multi-agent coordination.