OpenAI Codex CLI

Codex CLI is OpenAI's terminal-based coding agent. GROOVE integrates it as a provider for teams that want access to OpenAI's reasoning models alongside other tools.

Installation

npm i -g @openai/codex

Verify it's available:

codex --version

Authentication

Codex requires an OpenAI API key. Set it through the GUI's provider panel or via the CLI:

groove set-key codex sk-your-api-key

GROOVE encrypts the key with AES-256-GCM and stores it locally. The key is passed to the codex process at spawn time -- GROOVE never proxies API calls.

Models

Model	Tier	Best For
o3	Heavy	Complex multi-step reasoning, architecture decisions
o4-mini	Medium	General development, everyday coding tasks

When adaptive routing is set to Auto, GROOVE selects between these models based on task complexity classification.

GROOVE Integration

GROOVE spawns Codex as a managed process and monitors its stdout for status updates. Codex agents participate fully in the coordination layer -- they receive introductions, respect file locks, and route through the approval queue.

Limitations

• No hot-swap -- changing the model requires killing and respawning the agent. GROOVE handles this automatically during context rotation.
• API key billing -- usage is billed directly to your OpenAI account. Monitor costs through the Tokens tab in the GUI.

When to Use Codex

Codex is a good choice when you want OpenAI's reasoning models for specific tasks while running Claude Code as your primary provider. Spawn a Codex agent for a targeted subtask, let it finish, and kill it -- or let GROOVE's adaptive router decide when Codex is the right fit.