Docs

Codex

How LinkCode runs Codex, including sign-in, its always-autonomous approval behavior, model selection, and shared thread history.

Codex is bundled with LinkCode's host (the LinkCode daemon) — there is nothing to install.

Codex runs without asking

Codex threads in LinkCode execute commands and edit files without asking for approval — there is no approval policy to switch, unlike Claude Code. LinkCode also currently overrides your own Codex sandbox and approval settings with a more permissive configuration for every thread, regardless of what your own Codex config specifies. If that's not acceptable, don't run Codex on sensitive checkouts yet.

In development

A fix so LinkCode respects your own stricter Codex configuration, instead of overriding it, is in development. It will not add interactive per-action approvals — Codex's SDK has no callback for that, so prompting before each action isn't structurally possible today.

Sign in

LinkCode has no sign-in flow of its own for Codex. Two ways to authenticate:

  • Reuse an existing ChatGPT/Codex login on this machine.
  • Paste an API key into Settings → Agents. LinkCode uses it only for threads started with this agent.

If neither is set, Codex falls back to its own native sign-in resolution.

Model

Codex has no curated model list in LinkCode. Set one as free text in Settings → Agents' default-model field, using whatever model identifier Codex itself accepts. There's no reasoning-effort control for Codex — that field exists in Codex's own SDK, but LinkCode doesn't expose it yet.

Thread history

Resume is fully supported: stopping and later resuming a Codex thread works. History is shared with the standalone codex CLI — a thread started in either place reads back from the same on-disk store.

On this page