fix: derive service tier support from agent fast-mode metadata

api/config.py

@@ -2226,10 +2226,12 @@ def _apply_provider_prefix(
     result = []
     for m in raw_models:
         mid = m["id"]
+        entry = dict(m)
         if mid.startswith("@") or "/" in mid:
-            result.append({"id": mid, "label": m["label"]})
+            result.append(entry)
         else:
-            result.append({"id": f"@{provider_id}:{mid}", "label": m["label"]})
+            entry["id"] = f"@{provider_id}:{mid}"
+            result.append(entry)
     return result
 
 
@@ -3853,6 +3855,74 @@ def _is_openai_family_provider(provider: str | None) -> bool:
     return resolved in ("openai", "openai-api", "openai-codex")
 
 
+def _normalize_openai_family_model_id(model_id: str | None) -> str:
+    """Return a model id in the form expected by hermes_cli fast-mode resolution."""
+    model = str(model_id or "").strip()
+    if not model:
+        return ""
+
+    if model.startswith("@") and ":" in model:
+        model = model.split(":", 1)[1].strip()
+
+    if "://" in model:
+        return model
+
+    if "/" in model:
+        provider_hint, candidate = model.split("/", 1)
+        if provider_hint.strip().lower() in {"openai", "openai-api", "openai-codex"}:
+            model = candidate.strip()
+        else:
+            return ""
+
+    return model
+
+
+def _legacy_openai_service_tier_overrides(model_id: str | None, provider: str | None) -> dict:
+    """Compatibility fallback for standalone WebUI installs without hermes_cli.
+
+    Normal operation delegates to Hermes Agent model metadata.  This fallback
+    preserves the old WebUI behavior when the agent package is unavailable,
+    while still failing closed for codex model slugs and foreign provider IDs.
+    """
+    if not _is_openai_family_provider(provider):
+        return {}
+    resolved_provider = str(_resolve_provider_alias(str(provider or "").strip().lower()))
+    raw_model = str(model_id or "").strip()
+    if "://" not in raw_model and "/" in raw_model:
+        provider_hint = raw_model.split("/", 1)[0].strip().lower()
+        if provider_hint not in {"openai", "openai-api", "openai-codex"}:
+            return {}
+    normalized_model = _normalize_openai_family_model_id(model_id)
+    if not normalized_model:
+        if resolved_provider == "openai-codex":
+            return {}
+        return {"service_tier": "priority"}
+    lowered = normalized_model.lower()
+    if "codex" in lowered:
+        return {}
+    if lowered.startswith(("gpt-", "o1", "o3", "o4")):
+        return {"service_tier": "priority"}
+    return {}
+
+
+def _resolve_main_model_fast_mode_overrides(model_id: str | None, provider: str | None = None) -> dict:
+    """Return provider request overrides for the main model fast-mode setting."""
+    normalized_model = _normalize_openai_family_model_id(model_id)
+    if not normalized_model:
+        return _legacy_openai_service_tier_overrides(model_id, provider)
+    try:
+        from hermes_cli.models import resolve_fast_mode_overrides
+    except Exception:
+        logger.debug("Failed to import hermes_cli.models.resolve_fast_mode_overrides; using WebUI compatibility fallback.")
+        return _legacy_openai_service_tier_overrides(model_id, provider)
+    try:
+        resolved = resolve_fast_mode_overrides(normalized_model)
+    except Exception:
+        logger.debug("Failed to resolve fast-mode overrides for %r; using WebUI compatibility fallback.", normalized_model)
+        return _legacy_openai_service_tier_overrides(model_id, provider)
+    return resolved if isinstance(resolved, dict) else {}
+
+
 def _main_model_supports_service_tier(
     model_id: str | None,
     provider: str | None,
@@ -3860,26 +3930,43 @@ def _main_model_supports_service_tier(
     """Return True when the current main-model selection can use OpenAI service tier."""
     if not _is_openai_family_provider(provider):
         return False
-    resolved = str(provider or "").strip().lower()
-    if resolved == "openai-codex":
-        return False
-    raw_model = str(model_id or "").strip().lower()
-    if not raw_model:
-        return True
-    if "/" in raw_model:
-        prefix, bare_model = raw_model.split("/", 1)
-        if prefix != "openai":
-            return False
-    else:
-        bare_model = raw_model
-    if "codex" in bare_model:
-        return False
     return (
-        _is_first_party_model("openai", bare_model)
+        str(_resolve_main_model_fast_mode_overrides(model_id, provider).get("service_tier", "")).strip().lower()
-        or bare_model.startswith(("gpt-", "o1", "o3", "o4"))
+        == "priority"
     )
 
 
+def _model_supports_fast_tier_for_provider(model_id: str | None, provider: str | None) -> bool:
+    """Return whether a provider/model entry supports WebUI's service-tier toggle."""
+    return _main_model_supports_service_tier(model_id, provider)
+
+
+def _annotate_fast_tier_model_groups(payload: dict | None) -> dict | None:
+    """Add service-tier capability metadata to OpenAI-family model groups."""
+    if not isinstance(payload, dict):
+        return payload
+    groups = payload.get("groups")
+    if not isinstance(groups, list):
+        return payload
+    for group in groups:
+        if not isinstance(group, dict):
+            continue
+        provider_id = str(group.get("provider_id") or "").strip()
+        if not _is_openai_family_provider(provider_id):
+            continue
+        for bucket in ("models", "extra_models"):
+            models = group.get(bucket)
+            if not isinstance(models, list):
+                continue
+            for model in models:
+                if not isinstance(model, dict):
+                    continue
+                model_id = str(model.get("id") or "").strip()
+                if model_id:
+                    model["supports_fast_tier"] = _model_supports_fast_tier_for_provider(model_id, provider_id)
+    return payload
+
+
 def _public_main_service_tier(model_cfg: dict) -> str:
     """Return the saved main-model service tier only for OpenAI-family providers."""
     if not isinstance(model_cfg, dict):
@@ -4120,6 +4207,7 @@ def get_auxiliary_models() -> dict:
         "main": {
             "provider": main_provider,
             "model": main_model,
+            "supports_fast_tier": _main_model_supports_service_tier(main_model, main_provider),
             "service_tier": _public_main_service_tier(model_cfg),
             **_public_advanced_model_options(model_cfg),
         },
@@ -4447,13 +4535,13 @@ def _minimal_static_models_catalog() -> dict:
                     "models": [{"id": default_model, "label": label}],
                 }
             )
-        return {
+        return _annotate_fast_tier_model_groups({
             "active_provider": active_provider,
             "default_model": default_model,
             "configured_model_badges": {},
             "groups": groups,
             "aliases": {},
-        }
+        })
     except Exception:
         logger.debug("minimal static models catalog build failed", exc_info=True)
         return {
@@ -4812,7 +4900,7 @@ def _static_models_catalog_without_live_probes() -> dict:
         if not groups and default_model:
             return copy.deepcopy(_minimal_static_models_catalog())
 
-        return {
+        return _annotate_fast_tier_model_groups({
             "active_provider": active_provider,
             "default_model": default_model,
             "configured_model_badges": _configured_model_badges_from_static_catalog(
@@ -4822,7 +4910,7 @@ def _static_models_catalog_without_live_probes() -> dict:
             ),
             "groups": groups,
             "aliases": model_aliases,
-        }
+        })
     except Exception:
         logger.debug("static models catalog build failed", exc_info=True)
         return copy.deepcopy(_minimal_static_models_catalog())
@@ -5339,7 +5427,7 @@ def _load_models_cache_from_disk() -> dict | None:
         # disk save path does not persist `aliases`, so reconstruct them from
         # current config to keep the /api/models.aliases contract intact (a
         # disk-cache hit must not silently drop `/model <alias>` resolution).
-        return {
+        return _annotate_fast_tier_model_groups({
             "active_provider": cache["active_provider"],
             "default_model": cache["default_model"],
             "configured_model_badges": cache["configured_model_badges"],
@@ -5349,7 +5437,7 @@ def _load_models_cache_from_disk() -> dict | None:
                 if isinstance(cache.get("aliases"), dict)
                 else _model_aliases_from_config()
             ),
-        }
+        })
     except Exception:
         return None
 
@@ -5407,13 +5495,13 @@ def _load_stale_models_cache_from_disk() -> dict | None:
             # duration of the over-budget stale fallback. Reconstruct from
             # current config, mirroring the live/static catalog alias build.
             aliases = _model_aliases_from_config()
-        return {
+        return _annotate_fast_tier_model_groups({
             "active_provider": cache["active_provider"],
             "default_model": cache["default_model"],
             "configured_model_badges": cache["configured_model_badges"],
             "groups": cache["groups"],
             "aliases": aliases,
-        }
+        })
     except Exception:
         return None
 
@@ -5478,7 +5566,7 @@ def _get_fresh_memory_models_cache(now: float) -> dict | None:
         _available_models_cache_source_fingerprint = None
         return None
     if _is_valid_models_cache(_available_models_cache):
-        return copy.deepcopy(_available_models_cache)
+        return _annotate_fast_tier_model_groups(copy.deepcopy(_available_models_cache))
     _available_models_cache = None
     _available_models_cache_ts = 0.0
     _available_models_live_rebuild_ts = 0.0
@@ -6661,6 +6749,19 @@ def get_available_models(*, prefer_cache: bool = False, force_refresh: bool = Fa
                 allow_empty: bool = False,
             ) -> None:
                 picker_models = copy.deepcopy(raw_models or [])
+                if _is_openai_family_provider(provider_id):
+                    for _model in picker_models:
+                        if not isinstance(_model, dict):
+                            continue
+                        _model_id = str(_model.get("id") or "").strip()
+                        if not _model_id:
+                            continue
+                        _model["supports_fast_tier"] = (
+                            str(
+                                _resolve_main_model_fast_mode_overrides(_model_id, provider_id).get("service_tier", "")
+                            ).strip().lower()
+                            == "priority"
+                        )
                 if apply_prefix:
                     picker_models = _apply_provider_prefix(picker_models, provider_id, active_provider)
                 visible_models, extra_models = _split_picker_overflow_models(

api/routes.py

@@ -18552,7 +18552,11 @@ def _handle_live_models(handler, parsed):
         # Normalise to {id, label} — provider_model_ids() returns plain string IDs.
         # For ollama-cloud use the shared Ollama formatter (handles `:variant` suffix).
         # For all other providers use a simpler hyphen-split capitaliser.
-        from api.config import _format_ollama_label as _fmt_ollama
+        from api.config import (
+            _format_ollama_label as _fmt_ollama,
+            _is_openai_family_provider as _is_fast_tier_provider,
+            _model_supports_fast_tier_for_provider,
+        )
 
         def _make_label(mid):
             """Best-effort human label from a model ID string."""
@@ -18579,7 +18583,15 @@ def _handle_live_models(handler, parsed):
                 label = label.replace(orig.title(), orig)
             return label
 
-        models_out = [{"id": mid, "label": _make_label(mid)} for mid in ids if mid]
+        annotate_fast_tier = _is_fast_tier_provider(provider)
+        models_out = []
+        for mid in ids:
+            if not mid:
+                continue
+            entry = {"id": mid, "label": _make_label(mid)}
+            if annotate_fast_tier:
+                entry["supports_fast_tier"] = _model_supports_fast_tier_for_provider(mid, provider)
+            models_out.append(entry)
         return _finish({"provider": provider, "models": models_out,
                         "count": len(models_out)})
 

static/panels.js

@@ -9020,6 +9020,11 @@ async function loadSettingsPanel(){
           for(const m of [...(g.models||[]),...(g.extra_models||[])]){
             const opt=document.createElement('option');
             opt.value=m.id;opt.textContent=m.label;
+            if(m && (m.supports_fast_tier === true || String(m.supports_fast_tier).toLowerCase()==='true')){
+              opt.dataset.fast='1';
+            }else if(m && (m.supports_fast_tier === false || String(m.supports_fast_tier).toLowerCase()==='false')){
+              opt.dataset.fast='0';
+            }
             og.appendChild(opt);
           }
           modelSel.appendChild(og);
@@ -12045,17 +12050,9 @@ function _mainModelSupportsServiceTier(cfg){
  const optgroup=selectedOpt&&selectedOpt.parentElement&&selectedOpt.parentElement.tagName==='OPTGROUP'?selectedOpt.parentElement:null;
  const provider=((selectedOpt&&selectedOpt.dataset&&selectedOpt.dataset.provider)||(optgroup&&optgroup.dataset&&optgroup.dataset.provider)||(cfg&&cfg.provider)||'').trim().toLowerCase();
  if(provider!=='openai'&&provider!=='openai-api'&&provider!=='openai-codex') return false;
- if(provider==='openai-codex') return false;
+ const fastSupport = selectedOpt&&selectedOpt.dataset?selectedOpt.dataset.fast:'';
- const rawModel=String((selectedOpt&&selectedOpt.value)||(selected&&selected.value)||(cfg&&cfg.model)||'').trim().toLowerCase();
+ if(fastSupport) return fastSupport==='1'||fastSupport==='true';
- if(!rawModel) return true;
+ return cfg&&cfg.supports_fast_tier===true;
- let bareModel=rawModel;
- if(rawModel.includes('/')){
-  const slash=rawModel.indexOf('/');
-  if(rawModel.slice(0,slash)!=='openai') return false;
-  bareModel=rawModel.slice(slash+1);
- }
- if(bareModel.includes('codex')) return false;
- return bareModel.startsWith('gpt-')||bareModel.startsWith('o1')||bareModel.startsWith('o3')||bareModel.startsWith('o4');
 }
 
 function _openAuxAdvancedOptions(taskKey,cfg){

static/ui.js

@@ -3123,6 +3123,11 @@ async function populateModelDropdown(opts={}){
         const opt=document.createElement('option');
         opt.value=m.id;
         opt.textContent=m.label;
+        if(m && (m.supports_fast_tier === true || String(m.supports_fast_tier).toLowerCase()==='true')){
+          opt.dataset.fast='1';
+        }else if(m && (m.supports_fast_tier === false || String(m.supports_fast_tier).toLowerCase()==='false')){
+          opt.dataset.fast='0';
+        }
         og.appendChild(opt);
         _dynamicModelLabels[m.id]=m.label||m.id;
       }
@@ -3221,6 +3226,11 @@ function _addLiveModelsToSelect(provider, models, sel){
     opt.textContent=m.label||m.id;
     opt.title='Live model — fetched from provider';
     opt.dataset.provider=provider;
+    if(m && (m.supports_fast_tier === true || String(m.supports_fast_tier).toLowerCase()==='true')){
+      opt.dataset.fast='1';
+    }else if(m && (m.supports_fast_tier === false || String(m.supports_fast_tier).toLowerCase()==='false')){
+      opt.dataset.fast='0';
+    }
     providerGroup.appendChild(opt);
     _dynamicModelLabels[mid]=m.label||m.id;
     added++;

tests/test_auxiliary_models_settings.py

@@ -122,16 +122,19 @@ class TestAuxiliaryModelsJS:
         assert helper_idx >= 0
         helper_body = PANELS_JS[helper_idx:helper_idx + 1300]
         assert "_mainModelSupportsServiceTier" in PANELS_JS
-        assert "rawModel.includes('/')" in helper_body
+        assert "selectedOpt.dataset.fast" in helper_body
-        assert "rawModel.slice(0,slash)!=='openai'" in helper_body
+        assert "return cfg&&cfg.supports_fast_tier===true" in helper_body
-        assert "bareModel.includes('codex')" in helper_body
+        assert "return fastSupport==='1'||fastSupport==='true'" in helper_body
-        assert "bareModel.startsWith('gpt-')" in helper_body
+        assert "provider!=='openai'&&provider!=='openai-api'&&provider!=='openai-codex')" in helper_body
-        assert "bareModel.startsWith('o4')" in helper_body
+        assert "provider==='openai-codex') return false" not in helper_body
         assert "auxAdvancedServiceTier" in modal_body
         assert "isMain&&_mainModelSupportsServiceTier(cfg)" in modal_body
         assert "settings_main_advanced_service_tier" in modal_body
         assert "settings_main_advanced_service_tier_default" in modal_body
         assert "settings_main_advanced_service_tier_priority" in modal_body
+        assert "m.supports_fast_tier" in PANELS_JS
+        assert "opt.dataset.fast='1'" in PANELS_JS
+        assert "opt.dataset.fast='0'" in PANELS_JS
 
     def test_main_advanced_save_omits_unsupported_timing_keys(self):
         """Saving main-model options must not send blank timing keys that backend treats as clears."""

tests/test_credential_pool_providers.py

@@ -380,13 +380,14 @@ def test_apply_provider_prefix_no_double_prefix():
     from api.config import _apply_provider_prefix
 
     raw = [
-        {"id": "@copilot:gpt-5.4", "label": "already prefixed"},
+        {"id": "@copilot:gpt-5.4", "label": "already prefixed", "supports_fast_tier": True},
-        {"id": "openai/gpt-5.4", "label": "slash form"},
+        {"id": "openai/gpt-5.4", "label": "slash form", "supports_fast_tier": True},
-        {"id": "bare-model", "label": "bare"},
+        {"id": "bare-model", "label": "bare", "supports_fast_tier": False},
     ]
     result = _apply_provider_prefix(raw, "copilot", "openai-codex")
     ids = [m["id"] for m in result]
     assert ids == ["@copilot:gpt-5.4", "openai/gpt-5.4", "@copilot:bare-model"], ids
+    assert [m["supports_fast_tier"] for m in result] == [True, True, False]
 
 
 def test_apply_provider_prefix_active_provider_no_prefix():

tests/test_issue4536_service_tier.py

@@ -1,5 +1,7 @@
 """Tests for issue #4536 — main-model service_tier persistence and guarded forwarding."""
 
+import sys
+
 import pytest
 
 
@@ -61,6 +63,7 @@ class TestIssue4536ServiceTier:
 
         payload = config.get_auxiliary_models()
         assert payload["main"]["service_tier"] == "priority"
+        assert payload["main"]["supports_fast_tier"] is True
 
     def test_main_service_tier_default_clears_persisted_value(self, monkeypatch, tmp_path):
         """Choosing Default/off should clear service_tier from persisted main-model options."""
@@ -92,6 +95,11 @@ class TestIssue4536ServiceTier:
         )
         assert openai_payload == {"service_tier": "priority"}
 
+        codex_payload = config._main_model_request_overrides(
+            {"model": {"provider": "openai-codex", "default": "gpt-5.5", "service_tier": "priority"}},
+        )
+        assert codex_payload == {"service_tier": "priority"}
+
         openrouter_payload = config._main_model_request_overrides(
             {"model": {"provider": "openrouter", "default": "meta-llama/llama-3.1", "service_tier": "priority"}},
         )
@@ -107,10 +115,10 @@ class TestIssue4536ServiceTier:
         )
         assert openai_alias_payload == {"service_tier": "priority"}
 
-        codex_payload = config._main_model_request_overrides(
+        codex_nonfast_payload = config._main_model_request_overrides(
             {"model": {"provider": "openai-codex", "default": "gpt-5.3-codex", "service_tier": "priority"}},
         )
-        assert codex_payload == {}
+        assert codex_nonfast_payload == {}
 
         codex_empty_model_payload = config._main_model_request_overrides(
             {"model": {"provider": "openai-codex", "service_tier": "priority"}},
@@ -122,6 +130,16 @@ class TestIssue4536ServiceTier:
         )
         assert stale_openai_payload == {}
 
+        foreign_prefixed_openai_payload = config._main_model_request_overrides(
+            {"model": {"provider": "openai", "default": "openrouter/gpt-5.5", "service_tier": "priority"}},
+        )
+        assert foreign_prefixed_openai_payload == {}
+
+        codex_unknown_model_payload = config._main_model_request_overrides(
+            {"model": {"provider": "openai-codex", "default": "gpt-7.0-codex", "service_tier": "priority"}},
+        )
+        assert codex_unknown_model_payload == {}
+
     def test_auxiliary_payload_hides_service_tier_for_non_openai_main_models(self, monkeypatch, tmp_path):
         """Saved service_tier should not be re-exposed once the main model switches away from OpenAI."""
         from api import config
@@ -194,3 +212,57 @@ class TestIssue4536ServiceTier:
         assert result["ok"] is True
         text = config_path.read_text(encoding="utf-8")
         assert "service_tier" not in text
+
+    def test_standalone_agent_import_failure_preserves_supported_service_tier(self, monkeypatch, tmp_path):
+        """If hermes_cli is unavailable, Settings saves should not delete supported service_tier state."""
+        from api import config
+
+        config_path = tmp_path / "config.yaml"
+        config_path.write_text(
+            "model:\n  provider: openai-codex\n  default: gpt-5.5\n  service_tier: priority\n",
+            encoding="utf-8",
+        )
+        monkeypatch.setattr(config, "_get_config_path", lambda: config_path)
+        monkeypatch.setattr(config, "invalidate_models_cache", lambda: None)
+        monkeypatch.setattr(
+            config,
+            "cfg",
+            {"model": {"provider": "openai-codex", "default": "gpt-5.5", "service_tier": "priority"}},
+        )
+        monkeypatch.setitem(sys.modules, "hermes_cli.models", None)
+
+        assert config._main_model_request_overrides(
+            {"model": {"provider": "openai-codex", "default": "gpt-5.5", "service_tier": "priority"}},
+        ) == {"service_tier": "priority"}
+
+        result = config.set_hermes_default_model("gpt-5.5", provider="openai-codex")
+
+        assert result["ok"] is True
+        text = config_path.read_text(encoding="utf-8")
+        assert "service_tier: priority" in text
+
+    def test_standalone_agent_import_failure_still_blocks_codex_slug(self, monkeypatch):
+        """The compatibility fallback must not broadly enable Codex-specific model slugs."""
+        from api import config
+
+        monkeypatch.setitem(sys.modules, "hermes_cli.models", None)
+
+        assert config._main_model_request_overrides(
+            {"model": {"provider": "openai-codex", "default": "gpt-5.3-codex", "service_tier": "priority"}},
+        ) == {}
+
+    def test_auxiliary_payload_marks_false_fast_tier_for_unsupported_main_models(self, monkeypatch, tmp_path):
+        """The frontend needs explicit false metadata, not an absent field that falls back stale."""
+        from api import config
+
+        config_path = tmp_path / "config.yaml"
+        config_path.write_text(
+            "model:\n  provider: openai-codex\n  default: gpt-5.3-codex\n  service_tier: priority\n",
+            encoding="utf-8",
+        )
+        monkeypatch.setattr(config, "_get_config_path", lambda: config_path)
+
+        payload = config.get_auxiliary_models()
+
+        assert payload["main"]["service_tier"] == ""
+        assert payload["main"]["supports_fast_tier"] is False

tests/test_issue4756_session_visit_model_refresh.py

@@ -21,7 +21,7 @@ def _catalog(label: str) -> dict:
             {
                 "provider": "OpenAI",
                 "provider_id": "openai",
-                "models": [{"id": label, "label": label}],
+                "models": [{"id": label, "label": label, "supports_fast_tier": False}],
             }
         ],
         "aliases": {},

tests/test_live_models_ttl_cache.py

@@ -43,7 +43,7 @@ def test_live_models_cache_hits_within_ttl(monkeypatch):
 
     assert calls == ["openai"]
     assert first == second
-    assert first["models"] == [{"id": "openai/gpt-test", "label": "GPT Test"}]
+    assert first["models"] == [{"id": "openai/gpt-test", "label": "GPT Test", "supports_fast_tier": True}]
 
 
 def test_live_models_cache_expires(monkeypatch):
@@ -111,7 +111,7 @@ def test_live_models_cache_returns_deep_copies(monkeypatch):
     second = routes._handle_live_models(object(), parsed)
 
     assert second["provider"] == "openai"
-    assert second["models"] == [{"id": "openai/gpt-test", "label": "GPT Test"}]
+    assert second["models"] == [{"id": "openai/gpt-test", "label": "GPT Test", "supports_fast_tier": True}]
 
 
 def test_live_models_endpoint_respects_picker_visibility_budget(monkeypatch):

tests/test_model_cache_metadata.py

@@ -26,7 +26,7 @@ def test_save_models_cache_to_disk_preserves_response_metadata(tmp_path, monkeyp
             {
                 "provider": "OpenAI",
                 "provider_id": "openai",
-                "models": [{"id": "gpt-5.4-mini", "label": "GPT 5.4 Mini"}],
+                "models": [{"id": "gpt-5.4-mini", "label": "GPT 5.4 Mini", "supports_fast_tier": True}],
             }
         ],
     }