fix(model-routing): keep selected models strict (#5801)

Co-authored-by: Alexandre Teixeira <111787685+alteixeira20@users.noreply.github.com>

src/endpoint_resolver.py

@@ -443,28 +443,14 @@ def resolve_endpoint_by_id(
 
 
 def resolve_chat_fallback_candidates(owner: Optional[str] = None) -> list:
-    """Build the configured default-chat fallback chain as a list of
-    (chat_url, model, headers) tuples, skipping any that can't resolve.
+    """Compatibility shim for the retired default-chat fallback chain."""
 
-    The primary model is NOT included — callers prepend their session's
-    current (url, model, headers) so per-session model overrides are honored.
-    """
-    return _resolve_fallback_candidates("default_model_fallbacks", owner=owner)
+    del owner
+    return []
 
 
 def resolve_utility_fallback_candidates(owner: Optional[str] = None) -> list:
     """Configured fallback chain for the Utility model (`utility_model_fallbacks`)."""
-    try:
-        from src.settings import get_user_setting, load_settings
-        settings = load_settings()
-        utility_ep = (get_user_setting("utility_endpoint_id", owner or "", settings.get("utility_endpoint_id", "")) or "").strip()
-        if not utility_ep:
-            utility_chain = get_user_setting("utility_model_fallbacks", owner or "", settings.get("utility_model_fallbacks") or []) or []
-            if utility_chain:
-                return _resolve_fallback_candidates("utility_model_fallbacks", owner=owner)
-            return _resolve_fallback_candidates("default_model_fallbacks", owner=owner)
-    except Exception:
-        pass
     return _resolve_fallback_candidates("utility_model_fallbacks", owner=owner)
 
 

src/foreground_model_routing.py

@@ -0,0 +1,31 @@
+"""Foreground Chat and Agent model-routing policy.
+
+The selected session model is strict by default. Historical
+``default_model_fallbacks`` values remain stored for compatibility, but this
+policy intentionally does not read or migrate them.
+"""
+
+from typing import Any, Dict, Optional
+
+
+def resolve_foreground_fallback_candidates(owner: Optional[str] = None) -> list:
+    """Return fallback candidates for a foreground Chat or Agent request.
+
+    Foreground routing is strict, so no alternate endpoint/model is eligible.
+    ``owner`` is accepted to keep this policy boundary owner-aware.
+    """
+
+    del owner
+    return []
+
+
+def build_foreground_model_candidates(
+    endpoint_url: str,
+    model: str,
+    headers: Optional[Dict[str, Any]] = None,
+    owner: Optional[str] = None,
+) -> list:
+    """Build the ordered candidate list for a foreground request."""
+
+    primary = (endpoint_url, model, headers or {})
+    return [primary] + resolve_foreground_fallback_candidates(owner=owner)

src/llm_core.py

@@ -1949,11 +1949,10 @@ def _dedupe_candidates(candidates):
     """Filter malformed entries and drop a later repeat of an already-seen
     ``(url, model)`` route, preserving order (first occurrence wins).
 
-    The chain is the primary target followed by the configured fallbacks, so a
-    fallback that repeats the session's current model — a common misconfiguration,
-    since callers prepend the live ``(url, model)`` to ``default_model_fallbacks``
-    — would otherwise make the chain re-attempt the very route that just failed:
-    a wasted round-trip plus a spurious ``fallback`` notice for a switch that did
+    The chain is the primary target followed by any caller-authorized
+    fallbacks.  A fallback that repeats the session's current model would
+    otherwise make the chain re-attempt the very route that just failed: a
+    wasted round-trip plus a spurious ``fallback`` notice for a switch that did
     not happen. Headers are not part of the key; the first tuple (with its
     headers) is the one kept.
     """

src/settings.py

@@ -138,14 +138,13 @@ DEFAULT_SETTINGS = {
     # Email replies use email_writing_style instead because greetings,
     # signatures, and mailbox identity rules are medium-specific.
     "document_writing_style": "",
-    # Ordered fallback chain for the default chat model. Each entry is
-    # {"endpoint_id": "...", "model": "..."}. If the primary model fails
-    # before producing output (endpoint offline / errors), the chat
-    # dispatch retries the next entry in order.
+    # Legacy ordered fallback chain for the default chat model. Values remain
+    # stored for compatibility and rollback reference, but model routing no
+    # longer reads this key.
     "default_model_fallbacks": [],
-    # When True, non-admin users inherit global default model/endpoint/fallbacks
-    # when they have no personal defaults. When False, users only use their
-    # personal defaults (no global fallback). Default is False.
+    # When True, non-admin users inherit the global default model/endpoint when
+    # they have no personal defaults. When False, users only use their personal
+    # defaults. Default is False.
     "share_defaults_with_users": False,
     "utility_endpoint_id": "",
     "utility_model": "",

src/task_endpoint.py

@@ -32,7 +32,7 @@ def resolve_task_candidates(
     2. Utility endpoint/model
     3. Default endpoint/model
     4. Utility fallback chain
-    5. Default fallback chain
+    5. Retired default-fallback compatibility hook (currently empty)
     """
     candidates = []