Codex

Codex threads run through codex app-server, a long-lived process the Codex CLI provides for exactly this kind of integration. Cone spawns one shared server on demand and gives every Codex thread its own server-side thread on it. Codex requires CLI version 0.146 or newer.

How turns run

Starting a turn starts or resumes the thread on the app-server. Notifications stream back and render in the transcript the same way Claude turns do: reasoning, commands, file changes, and token usage as they happen.

Because the server connection stays open, Codex threads get the full interactive flow locally:

• Approval requests appear as permission cards in the thread that caused them, scoped by the thread’s access mode.
• Stop interrupts only the current thread’s turn. Other Codex threads and the shared server keep running.
• Thread state resumes across app relaunches through the stored thread id.

Models and effort

Cone loads the model catalog live from the CLI, so new models appear without an app update. Models expose their reasoning-effort options in the picker. Fast mode maps to the model’s priority service tier where one exists.

If the codex CLI is missing or too old, existing Codex threads stay readable and their stored models render as placeholders until the CLI is back.

Current limits

Codex sessions are local only. SSH remotes and agent profiles are currently Claude features.