fix(hermes_cli): keep TTS/setup xAI OAuth from switching active chat provider

Save side-tool OAuth tokens without promoting xai-oauth via active_provider
or model.provider so hermes setup tts login no longer hijacks inference routing.

hermes_cli/auth.py

@@ -4452,7 +4452,15 @@ def _save_xai_oauth_tokens(
     redirect_uri: str = "",
     last_refresh: Optional[str] = None,
     auth_mode: str = "oauth_device_code",
+    set_active: bool = True,
 ) -> None:
+    """Persist xAI OAuth tokens into the auth store.
+
+    When *set_active* is True (default), also promote ``xai-oauth`` to
+    ``active_provider`` — appropriate for intentional model/auth login.
+    Pass ``set_active=False`` for side-tool credential bootstrap (TTS, STT,
+    dashboard token save, token refresh) so inference routing is unchanged.
+    """
     if last_refresh is None:
         last_refresh = datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
     with _auth_store_lock():
@@ -4470,7 +4478,9 @@ def _save_xai_oauth_tokens(
             state["discovery"] = discovery
         if redirect_uri:
             state["redirect_uri"] = redirect_uri
-        _save_provider_state(auth_store, "xai-oauth", state)
+        _store_provider_state(
+            auth_store, "xai-oauth", state, set_active=set_active
+        )
         _save_auth_store(auth_store)
         if write_through_to_root:
             _write_through_xai_oauth_to_global_root(state)
@@ -4808,6 +4818,9 @@ def _refresh_xai_oauth_tokens(
         redirect_uri=redirect_uri,
         last_refresh=refreshed["last_refresh"],
         auth_mode=auth_mode,
+        # Refresh must not flip active_provider — TTS/side tools can refresh
+        # xAI tokens while chat still routes through another provider.
+        set_active=False,
     )
     return updated_tokens
 

hermes_cli/setup.py

@@ -896,15 +896,17 @@ def _xai_oauth_logged_in_for_setup() -> bool:
 def _run_xai_oauth_login_from_setup() -> bool:
     """Run the xAI Grok OAuth device-code login from inside the setup wizard.
 
+    Saves OAuth tokens only. Does **not** switch the active inference
+    provider or rewrite ``model.provider`` — callers (TTS setup, tools
+    config) only need credentials for side tools.
+
     Returns True on success, False on any failure (the caller falls back
     to whatever the user picked next, e.g. Edge TTS).
     """
     try:
         from hermes_cli.auth import (
-            DEFAULT_XAI_OAUTH_BASE_URL,
             _is_remote_session,
             _save_xai_oauth_tokens,
-            _update_config_for_provider,
             _xai_oauth_device_code_login,
         )
     except Exception as exc:
@@ -922,9 +924,7 @@ def _run_xai_oauth_login_from_setup() -> bool:
             redirect_uri=creds.get("redirect_uri", ""),
             last_refresh=creds.get("last_refresh"),
             auth_mode="oauth_device_code",
-        )
-        _update_config_for_provider(
-            "xai-oauth", creds.get("base_url", DEFAULT_XAI_OAUTH_BASE_URL)
+            set_active=False,
         )
         return True
     except Exception as exc:

hermes_cli/web_server.py

@@ -11153,6 +11153,9 @@ def _xai_device_poller(session_id: str) -> None:
                 discovery=discovery,
                 last_refresh=datetime.now(timezone.utc).isoformat().replace("+00:00", "Z"),
                 auth_mode="oauth_device_code",
+                # Dashboard OAuth only bootstraps credentials for side tools;
+                # do not hijack the active chat inference provider.
+                set_active=False,
             )
             # The singleton write above is the single source of truth: the
             # credential-pool load seeds it as the canonical ``device_code``