OpenAI Codex
How AgentDock detects, launches, and integrates with the OpenAI Codex CLI.
OpenAI Codex is one of the three coding-agent CLIs AgentDock can drive. As with the other agents, you install and authenticate the codex CLI yourself — AgentDock only detects and launches whatever is already set up on your system.
Prerequisites
- The Codex CLI installed and authenticated, per OpenAI's own instructions.
- No AgentDock-specific setup beyond that — see Getting started: connecting your agents.
How AgentDock detects it
Codex detection is different from Claude Code and Antigravity. Instead of a plain PATH search, AgentDock uses a multi-tier resolver, checked in order:
- A custom executable path override, if you've set one in Settings.
- The Codex SDK's own bundled native runtime.
- A standalone
codexinstallation. - If none of those resolve, detection fails with a specific error message.
Why Codex is special-cased
On Windows, a bare codex on PATH commonly resolves to an npm .cmd shim, which the Codex SDK's internal spawn call can't launch directly. The tiered resolver exists specifically to avoid that failure mode rather than relying on a plain PATH lookup like the other two agents.
You can still override the resolved path in Settings → Agents, which runs a live "Test" action and shows diagnostics.
How the integration works
AgentDock talks to Codex through @openai/codex-sdk. The model catalogue is fetched live from the CLI's own app-server, so the models you can pick from reflect what your installed Codex CLI actually supports.
Sandbox/permission mode is set when a Codex thread is created and does not switch live mid-session the way Claude's does — see Permissions for the full list of modes (default, read-only, workspace-write, danger-full-access, bypass).