diff --git a/veadk/runtime/__init__.py b/veadk/runtime/__init__.py
index d5c5a66f3..350e10cdd 100644
--- a/veadk/runtime/__init__.py
+++ b/veadk/runtime/__init__.py
@@ -61,8 +61,8 @@ def get_runtime(name: str) -> BaseRuntime:
         except ModuleNotFoundError as e:
             raise ImportError(
                 f"The 'codex' runtime requires extra dependencies (missing: {e.name}). "
-                "Install the Codex binary and the codex_app_server SDK, plus: "
-                "pip install fastapi uvicorn"
+                "Install them with: pip install openai-codex fastapi uvicorn "
+                "(openai-codex bundles the Codex binary via openai-codex-cli-bin)."
             ) from e
 
         return CodexRuntime()
diff --git a/veadk/runtime/codex/runtime.py b/veadk/runtime/codex/runtime.py
index c6e066430..4bd835c4b 100644
--- a/veadk/runtime/codex/runtime.py
+++ b/veadk/runtime/codex/runtime.py
@@ -28,7 +28,8 @@
 - Codex only speaks the Responses API, so requests are routed through an
   in-process Responses→chat shim (see :mod:`veadk.runtime.codex.proxy`).
 
-Note: this requires the Codex binary on PATH and the ``codex_app_server`` SDK.
+Note: this requires the ``openai-codex`` SDK (``pip install openai-codex``),
+which bundles the Codex CLI binary via its ``openai-codex-cli-bin`` dependency.
 """
 
 from __future__ import annotations
@@ -37,7 +38,7 @@
 import tempfile
 from typing import TYPE_CHECKING, AsyncGenerator
 
-from codex_app_server import AsyncCodex  # type: ignore[import-not-found]
+from openai_codex import AsyncCodex  # type: ignore[import-not-found]
 
 from veadk.runtime.base_runtime import BaseRuntime, build_system_append
 from veadk.runtime.codex.proxy import get_shim_url