Merge remote-tracking branch 'origin/main' into ethie/pm-clean

gateway/run_agent_cache.py

@@ -160,10 +160,13 @@ class GatewayAgentCacheMixin:
             return
         override: Dict[str, Any] = {k: persisted.get(k) for k in ("model", "provider", "base_url")}
         provider = persisted.get("provider")
+        from hermes_cli.runtime_provider import is_foreign_provider_endpoint
+        if is_foreign_provider_endpoint(provider, override.get("base_url")):
+            override["base_url"] = None  # left over from a switch that kept the previous provider's URL
         if provider:
             # Re-resolve credentials for the persisted provider. On failure (e.g. credentials removed
             # since the switch) keep the credential-less override — _resolve_session_agent_runtime
-            # falls back to env resolution and layers model/provider.
+            # retries the resolution for that provider on each turn (default route + notice meanwhile).
             try:
                 runtime = _resolve_runtime_agent_kwargs_for_provider(provider, target_model=persisted.get("model") or None)
                 for k in ("api_key", "api_mode", "credential_pool", "requested_provider", "max_tokens"):

gateway/run_turn.py

@@ -207,7 +207,8 @@ class GatewayTurnMixin:
                     skey or "", model, override_model, override_runtime.get("provider"),
                 )
                 return override_model, override_runtime
-            # No api_key on the override: env-based resolution below, override model/provider on top.
+            # No api_key on the override (credentials failed to re-resolve at rehydrate): resolve them
+            # for the override's own provider below, never layer it over the default provider's runtime.
             logger.debug(
                 "Session model override (no api_key, fallback): session=%s config_model=%s override_model=%s",
                 skey or "", model, override_model,
@@ -222,7 +223,19 @@ class GatewayTurnMixin:
                 ][:5] or "[]",
             )
 
-        runtime_kwargs = _resolve_runtime_agent_kwargs()
+        runtime_kwargs, unavailable_override = None, None
+        if override and override.get("provider"):
+            try:
+                runtime_kwargs = _resolve_runtime_agent_kwargs_for_provider(
+                    override["provider"], target_model=override.get("model") or None)
+            except Exception as exc:
+                # Layering the override on the default runtime sent its model to the default provider's
+                # endpoint (openai-codex on the Nous URL). Run this turn on the whole default route and say
+                # so; the persisted override is kept, so the next turn retries it.
+                logger.warning("Session /model override provider %s unavailable: %s", override["provider"], exc)
+                unavailable_override, override = override, None
+        if runtime_kwargs is None:
+            runtime_kwargs = _resolve_runtime_agent_kwargs()
         # Private notice metadata must never reach an ``AIAgent(**runtime_kwargs)`` spread; the turn
         # runner surfaces it through the agent's one-shot fallback notice (#74349).
         self._pre_agent_fallback_notice = runtime_kwargs.pop("_fallback_notice", None)
@@ -230,6 +243,10 @@ class GatewayTurnMixin:
         if runtime_model:
             logger.info("Runtime provider supplied explicit model override: %s -> %s", model, runtime_model)
             model = runtime_model
+        if unavailable_override and not self._pre_agent_fallback_notice:
+            from hermes_cli.fallback_config import pre_agent_fallback_notice
+            self._pre_agent_fallback_notice = pre_agent_fallback_notice(
+                unavailable_override["provider"], unavailable_override.get("model"), runtime_kwargs.get("provider"), model)
 
         cfg = getattr(self, "config", None)  # getattr: bare object.__new__ test runners
         if cfg and source is not None:

gateway/slash_commands_model.py

@@ -215,6 +215,7 @@ class GatewayModelCommandsMixin:
                     _sess_entry.was_auto_reset = False
                 await _sess_db.update_session_model(
                     _sess_entry.session_id, result.new_model, provider=result.target_provider,
+                    base_url=result.base_url, api_mode=result.api_mode,
                 )
             except Exception as exc:
                 logger.debug("Failed to persist model switch to DB: %s", exc)