Codex Preferences

Open Preferences -> AI Integration -> Codex. Enable Codex integration (will be available in Graph popup menus) makes Codex run actions available.

• Path selects the installed Codex executable or a custom launcher. You can also enter a command name resolved from PATH.
• Arguments supplies the launch arguments for both new and resumed conversations. The Variables menu inserts launcher placeholders.
• Run Codex in a dedicated SmartGit worktree controls worktree isolation.
• Model and Reasoning Effort, under Defaults, set the initial model settings for conversations. Choose from the suggestions or enter values supported by your installed agent and model.
• Service Tier offers <default>, auto, flex, and priority, or a custom value. <default> leaves Codex’s service-tier choice unchanged.
• Context Window accepts a positive token count. <default> uses the model’s default.
• Sandboxing selects the permissions described under Sandboxing.

Auto-Configure replaces the launch settings and restores the model defaults. It also resets Service Tier and Context Window to <default> and clears Working Directory and Cleanup Arguments. To apply the detected setup, select I understand that Codex will run without interactive permission prompts and click Allow and Apply.

Custom Codex launchers

The Custom containerized Codex fields are intended for custom launchers:

• Working Directory sets the launcher’s working directory. Leave it empty to use the execution worktree. Restricted sandbox profiles always use the execution worktree.
• Cleanup Arguments optionally invokes the same launcher to clean up a session’s environment when the session is deleted.