fix(agent): validation robuste des modeles Nvidia (payload production + repli 404/410) + logo noir/blanc

Validation: - _validate_chat_models utilise le payload de production (temperature 0.2, sans max_tokens) : un 200 a la validation == 200 a l'usage reel. - Race corrigee : payload reconstruit par modele (plus de noms croises entre requetes concurrentes) — la v5.1.8 retombait par intermittence sur la liste brute (68 modeles dont la plupart en 404/410). - Retente une fois sur timeout/5xx (les modeles lents mais fonctionnels survivent) ; rejette les 4xx (404 inconnu, 410 retire) ; garde les 429. - Verifie en live contre l'API NVIDIA : 12 modeles valides au lieu de 68. Repli runtime: - LLMClient._http_complete : sur 404/410, retente une fois avec le modele par defaut du provider et marque un notice dans LLMResponse. - AgentEngine emet un evenement SSE "notice" (bandeau .fd-ap-notice dans le panneau Agent, reset a chaque conversation) ; le modele reel est persistee. UI: - FAB (rond bas droite) et logo header du panneau : rond noir/blanc qui suit le theme clair/sombre (--text-primary / --bg-primary), eclair monochrome SVG a la place de l'emoji robot. Tests: - 2 nouveaux tests (429 garde / 410 retire ; repli runtime sur 410). - 278 tests passent.
2026-09-07 14:45:28 -04:00
parent b15289b4bc
commit bd582866d1
8 changed files with 215 additions and 76 deletions
@@ -1,5 +1,40 @@
 # Changelog — FlowDeck

+## v5.1.9 (2026-09-07) — Validation robuste des modèles Nvidia + logo Agent noir & blanc
+
+> La liste Nvidia « diminuée » de la v5.1.8 contenait encore des modèles en
+> erreur 404/410. Deux causes réelles : (1) le payload de validation ne
+> correspondait pas à celui utilisé au runtime et un `payload` partagé entre
+> les requêtes concurrentes envoyait des noms de modèles croisés (race) ;
+> (2) rien ne protégeait l'usage si un modèle devenait indisponible après
+> validation.
+
+### Validation des modèles (fournisseur bruité)
+- `_validate_chat_models()` utilise désormais le **payload de production**
+  (`messages` + `temperature: 0.2`, sans `max_tokens`) : un modèle qui répond
+  200 à la validation répond aussi 200 à l'usage réel.
+- **Race fixée** : le payload est reconstruit pour chaque modèle (plus de nom
+  de modèle croisé entre requêtes concurrentes) — c'est pourquoi la v5.1.8
+  retombait par intermittence sur la liste brute non validée.
+- Un seul payload partagé retiré ; retente **une fois** sur timeout/5xx (les
+  modèles lents mais fonctionnels survivent) ; rejette définitivement les 4xx
+  (404 modèle inconnu, 410 retiré/déprécié) ; garde les 429 (routés donc
+  utilisables).
+
+### Repli automatique au runtime (model not found / gone)
+- `LLMClient._http_complete` : si le modèle choisi répond **404/410**, retente
+  une fois avec le modèle par défaut du provider et ajoute un `notice`
+  expliquant le basculement — visible dans le panneau Agent (nouvel événement
+  SSE « notice », bandeau `.fd-ap-notice`, réinitialisé à chaque conversation).
+- Le modèle réellement utilisé est persisté dans l'historique de conversation.
+
+### Panneau Agent — logo noir & blanc thémé
+- Le rond bleu en bas à droite (FAB) et le logo du header du panneau (à gauche
+  de « How can I help you today? ») passent de l'accent bleu + émoji 🤖 à un
+  rond noir/blanc qui suit le thème clair/sombre, avec une étincelle ✨
+  monochrome (SVG `currentColor` : rond `--text-primary`, étincelle
+  `--bg-primary`).
+
 ## v5.1.8 (2026-09-07) — Multi-lignes réellement conservées + modèles IA validés (Nvidia)

 > Deux correctifs : le bug multi-lignes (v5.1.7 corrigeait le symptôme au
@@ -1 +1 @@
-5.1.8
+5.1.9
@@ -60,7 +60,7 @@ async def lifespan(_app: FastAPI):

 app = FastAPI(
    title="FlowDeck",
-    version="5.1.8",
+    version="5.1.9",
    docs_url="/docs" if settings.log_level == "DEBUG" else None,
    redoc_url=None,
    lifespan=lifespan,
@@ -154,6 +154,7 @@ class AgentEngine:

        tool_schema = self.tools.schema(scope)
        final_text = None
+        used_model = model or ""  # peut être ajusté par un repli de modèle (404/410)

        try:
            for step in range(settings.agent_max_iterations or MAX_ITERATIONS):
@@ -167,6 +168,11 @@ class AgentEngine:
                )
                self._tokens += response.usage.get("total_tokens", 0) or 0

+                if getattr(response, "notice", ""):
+                    yield self._event("notice", {"message": response.notice})
+
+                used_model = getattr(response, "model", "") or used_model
+
                if response.text and response.text.strip():
                    yield self._event("reasoning", {"content": response.text})

@@ -242,7 +248,7 @@ class AgentEngine:
                yield self._event("final", {"content": final_text})

            self._persist_message(conversation_id, "assistant", final_text,
-                                  model=model, tokens=self._tokens)
+                                  model=used_model, tokens=self._tokens)
            await self._autotitle(conversation_id, objective, final_text)

        except Exception as exc:  # noqa: BLE001
@@ -85,6 +85,7 @@ class LLMResponse:
    tool_calls: list[dict] = field(default_factory=list)
    model: str = ""
    usage: dict = field(default_factory=dict)
+    notice: str = ""


 class LLMClient:
@@ -161,7 +162,7 @@ class LLMClient:
    def _endpoint(self) -> str:
        return f"{self.api_base.rstrip('/')}/chat/completions"

-    async def _http_complete(self, messages, model, tools) -> LLMResponse:
+    async def _http_complete(self, messages, model, tools, *, _noticer: str = "") -> LLMResponse:
        payload: dict = {
            "model": model,
            "messages": messages,
@@ -173,11 +174,34 @@ class LLMClient:
        headers = {"Content-Type": "application/json"}
        if self.api_key:
            headers["Authorization"] = f"Bearer {self.api_key}"
-        async with httpx.AsyncClient(timeout=settings.agent_run_timeout_seconds) as client:
-            resp = await client.post(self._endpoint(), json=payload, headers=headers)
-            resp.raise_for_status()
-            data = resp.json()
+        try:
+            async with httpx.AsyncClient(timeout=settings.agent_run_timeout_seconds) as client:
+                resp = await client.post(self._endpoint(), json=payload, headers=headers)
+                resp.raise_for_status()
+                data = resp.json()
+        except httpx.HTTPStatusError as exc:
+            # Repli robuste : le modèle choisi a été retiré / n'existe plus
+            # (404 « model not found » / 410 « has reached its end of life »).
+            # Au lieu d'échouer, on retente UNE fois avec le modèle par défaut du
+            # provider et on signale le basculement — la liste validée peut avoir
+            # vieilli (modèle déprécié entre deux rafraîchissements).
+            if exc.response.status_code in (404, 410) \
+                    and model and self.default_model and model != self.default_model:
+                logger.warning(
+                    "Model '%s' unavailable (%s) on %s — retrying with default '%s'",
+                    model, exc.response.status_code, self.provider, self.default_model,
+                )
+                return await self._http_complete(messages, self.default_model, tools, _noticer=(
+                    f"Le modèle « {model} » n'est plus disponible ({exc.response.status_code}). "
+                    f"Réponse générée avec « {self.default_model} » à la place."
+                ))
+            raise

+        response = self._parse_response(data, model)
+        response.notice = _noticer or ""
+        return response
+
+    def _parse_response(self, data: dict, model: str) -> LLMResponse:
        choice = data["choices"][0]["message"]
        text = choice.get("content") or ""
        tool_calls = []
@@ -193,7 +217,6 @@ class LLMClient:
                "arguments": args,
                "arguments_raw": fn.get("arguments") or "",
            })
-
        return LLMResponse(
            text=text,
            tool_calls=tool_calls,
@@ -351,44 +351,52 @@ async def fetch_provider_models(provider: str, *, api_key: str = "",


 async def _validate_chat_models(base_url: str, api_key: str, candidates: list[str],
-                                *, timeout: float = 6.0, concurrency: int = 6,
-                                deadline: float = 55.0) -> list[str]:
-    """Probe `POST {base_url}/chat/completions` for each candidate with a minimal
-    payload and keep the models that accept it. A 429 (rate limit) is treated as
-    "probably fine" since it proves the model is routed, not that it's invalid."""
+                                *, timeout: float = 12.0, concurrency: int = 5,
+                                deadline: float = 90.0, attempts: int = 2) -> list[str]:
+    """Probe `POST {base_url}/chat/completions` for each candidate and keep only
+    the models that genuinely answer — using the exact payload the app sends at
+    runtime (`temperature: 0.2`, no `max_tokens`).
+
+    Hard failures (404 model inconnu, 410 modèle retiré…) exclude the model.
+    A 429 (rate limit) is kept: it proves the route exists, so the model is
+    usable once the quota frees up. Transient failures (timeouts, 5xx, network)
+    are retried once before giving up, so slow-but-working models survive.
+    """
    import asyncio
    import httpx

-    valid: list[str] = []
    sem = asyncio.Semaphore(concurrency)
    start = time.monotonic()
    headers = {"Content-Type": "application/json"}
    if api_key:
        headers["Authorization"] = f"Bearer {api_key}"
-    payload: dict = {
-        "model": None,
+    url = f"{base_url.rstrip('/')}/chat/completions"
+    payload_tpl = {
        "messages": [{"role": "user", "content": "ping"}],
-        "max_tokens": 4,
+        "temperature": 0.2,
    }

    async def probe(model: str) -> str | None:
        if time.monotonic() - start > deadline:
            return None
-        payload["model"] = model
-        try:
-            async with sem:
-                async with httpx.AsyncClient(timeout=timeout) as client:
-                    resp = await client.post(
-                        f"{base_url}/chat/completions", headers=headers, json=payload
-                    )
-            if resp.status_code < 300 or resp.status_code == 429:
-                return model
-        except Exception:  # noqa: BLE001 — réseau/timeout ⇒ rejeté
-            return None
+        payload: dict = {"model": model, **payload_tpl}
+        for attempt in range(attempts):
+            if time.monotonic() - start > deadline:
+                return None
+            try:
+                async with sem:
+                    async with httpx.AsyncClient(timeout=timeout) as client:
+                        resp = await client.post(url, headers=headers, json=payload)
+                code = resp.status_code
+                if code < 300 or code == 429:
+                    return model
+                if code < 500:  # 400/401/403/404/410… → définitivement non utilisable
+                    return None
+                # 5xx → passage passager, on retente une fois
+            except Exception:  # noqa: BLE001 — timeouts / connexion → on retente
+                if attempt >= attempts - 1:
+                    return None
        return None

    results = await asyncio.gather(*(probe(m) for m in candidates))
-    for m in results:
-        if m:
-            valid.append(m)
-    return valid
+    return [m for m in results if isinstance(m, str)]
@@ -1,7 +1,7 @@
 {# ════════════ FlowDeck Agent panel v4.13.0 — Notion-AI-style sidebar ════════════ #}
 <style>
-#fd-agent-fab{position:fixed;right:22px;bottom:22px;width:54px;height:54px;border-radius:50%;background:var(--accent,#2383E2);color:#fff;border:none;font-size:24px;cursor:pointer;box-shadow:0 8px 26px rgba(35,131,226,.5);display:flex;align-items:center;justify-content:center;z-index:1700;transition:transform .15s,box-shadow .15s}
-#fd-agent-fab:hover{transform:scale(1.06);box-shadow:0 12px 34px rgba(35,131,226,.62)}
+#fd-agent-fab{position:fixed;right:22px;bottom:22px;width:54px;height:54px;border-radius:50%;background:var(--text-primary,#37352F);color:var(--bg-primary,#fff);border:none;cursor:pointer;box-shadow:0 8px 26px rgba(0,0,0,.35);display:flex;align-items:center;justify-content:center;z-index:1700;transition:transform .15s,box-shadow .15s}
+#fd-agent-fab:hover{transform:scale(1.06);box-shadow:0 12px 34px rgba(0,0,0,.45)}
 #fd-agent-panel{position:fixed;top:0;right:0;bottom:0;width:470px;max-width:100vw;background:var(--bg-primary,#191919);border-left:1px solid var(--border,rgba(255,255,255,.1));z-index:1850;display:flex;flex-direction:column;box-shadow:-16px 0 48px rgba(0,0,0,.55);transform:translateX(102%);transition:transform .24s cubic-bezier(.16,1,.3,1)}
 #fd-agent-panel.open{transform:translateX(0)}
 #fd-agent-panel ::-webkit-scrollbar{width:8px;height:8px}
@@ -10,7 +10,7 @@

 /* Header */
 .fd-ap-head{display:flex;align-items:center;gap:12px;padding:16px 18px 12px}
-.fd-ap-logo{width:36px;height:36px;border-radius:50%;background:linear-gradient(135deg,var(--accent,#2383E2),#5b8def);color:#fff;display:flex;align-items:center;justify-content:center;font-size:19px;flex-shrink:0;box-shadow:0 4px 12px rgba(35,131,226,.35)}
+.fd-ap-logo{width:36px;height:36px;border-radius:50%;background:var(--text-primary,#37352F);color:var(--bg-primary,#fff);display:flex;align-items:center;justify-content:center;flex-shrink:0;box-shadow:0 2px 8px rgba(0,0,0,.25)}
 .fd-ap-head-main{flex:1;min-width:0}
 .fd-ap-title{font-size:15px;font-weight:600;color:var(--text,#f0f0f0);white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
 .fd-ap-sub{font-size:11.5px;color:var(--text-dim,#9aa);margin-top:2px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}
@@ -64,6 +64,7 @@
 .fd-proposal-note{font-size:11px;color:var(--text-dim,#999)}
 .fd-agent-thinking{font-size:12.5px;color:var(--text-dim,#999);padding:10px 12px;border-radius:9px;background:rgba(255,255,255,.04);margin:4px 0 8px;font-style:italic}
 .fd-ap-err{color:#ef8a8a;font-size:12px;padding:4px 2px}
+.fd-ap-notice{color:var(--text-secondary,#ccc);font-size:12px;padding:7px 10px;margin:2px 0 8px;border-radius:9px;background:rgba(128,128,128,.12);border:1px solid var(--border,rgba(255,255,255,.12))}

 /* Composer */
 .fd-ap-composer{padding:10px 14px 14px;border-top:1px solid var(--border,rgba(255,255,255,.09));position:relative}
@@ -191,12 +192,12 @@
 </script>

 {# ── Floating button ── #}
-<button id="fd-agent-fab" title="FlowDeck Agent (Ctrl+J)" onclick="window.fdAgent&&window.fdAgent.toggle()">🤖</button>
+<button id="fd-agent-fab" title="FlowDeck Agent (Ctrl+J)" onclick="window.fdAgent&&window.fdAgent.toggle()"><svg viewBox="0 0 24 24" width="26" height="26" fill="currentColor" aria-hidden="true"><path d="M12 3l1.9 5.8a2 2 0 0 0 1.3 1.3L21 12l-5.8 1.9a2 2 0 0 0-1.3 1.3L12 21l-1.9-5.8a2 2 0 0 0-1.3-1.3L3 12l5.8-1.9a2 2 0 0 0 1.3-1.3L12 3Z"/></svg></button>

 {# ── Panel ── #}
 <div id="fd-agent-panel" class="fd-ap" :class="{open: open}" x-data="agentPanel()" @keydown.escape.window="close()" role="dialog" aria-label="Agent FlowDeck">
  <div class="fd-ap-head">
-    <span class="fd-ap-logo">🤖</span>
+    <span class="fd-ap-logo"><svg viewBox="0 0 24 24" width="19" height="19" fill="currentColor" aria-hidden="true"><path d="M12 3l1.9 5.8a2 2 0 0 0 1.3 1.3L21 12l-5.8 1.9a2 2 0 0 0-1.3 1.3L12 21l-1.9-5.8a2 2 0 0 0-1.3-1.3L3 12l5.8-1.9a2 2 0 0 0 1.3-1.3L12 3Z"/></svg></span>
    <div class="fd-ap-head-main">
      <div class="fd-ap-title">How can I help you today?</div>
      <div class="fd-ap-sub" x-text="providerModelLabel"></div>
@@ -276,6 +277,7 @@
          </div>
        </template>
        <div class="fd-agent-thinking" x-show="streaming">L'agent réfléchit et agit… <span x-text="liveReasoning"></span></div>
+        <div class="fd-ap-notice" x-show="notice" x-cloak x-text="notice"></div>
        <div class="fd-ap-err" x-show="errorMsg" x-text="errorMsg"></div>
      </div>
    </template>
@@ -412,7 +414,7 @@
      open: false, modelOpen: false, tab: 'chat',
      agents: [], conversations: [], currentConv: null,
      messages: [], input: '', streaming: false, liveReasoning: '',
-      provider: 'offline', providerAvailable: false, errorMsg: '',
+      provider: 'offline', providerAvailable: false, errorMsg: '', notice: '',
      allProviders: [], providers: [], providerModels: [], llmProvider: 'offline',
      llmModel: '', llmKeys: [], defaultProvider: 'offline',
      activeContext: null, ctxPinnedKey: null,
@@ -756,7 +758,7 @@
          .then(function(r){return r.json()}).then(function(d){
            var conv = {id:d.id, title:d.title, updated_at:new Date().toISOString()};
            self.conversations.unshift(conv); self.currentConv = conv;
-            self.messages = []; self.errorMsg=''; self.tab='chat';
+            self.messages = []; self.errorMsg=''; self.notice=''; self.tab='chat';
            self.contextChips = []; self.mentionOpen = false;
            self.$nextTick(function(){ self.resizeInput(); });
          }).catch(function(){});
@@ -773,7 +775,7 @@
        }).then(function(r){return r.json()}).then(function(d){
          var conv = {id:d.id, title:d.title, updated_at:new Date().toISOString()};
          self.conversations.unshift(conv); self.currentConv = conv;
-          self.messages = []; self.errorMsg=''; self.tab='chat';
+          self.messages = []; self.errorMsg=''; self.notice=''; self.tab='chat';
          self.contextChips = []; self.mentionOpen = false;
          self._ensuring = null;
          return conv;
@@ -1682,6 +1684,7 @@
                    self.messages.push({id:'f-'+Date.now(), role:'assistant', content: ev.content});
                    if(window.__fdInsertAI){ try{ var h=window.__fdInsertAI; window.__fdInsertAI=null; h(ev.content||''); }catch(e){} }
                  }
+                  else if(ev.type==='notice'){ self.notice = ev.message; }
                  else if(ev.type==='error'){ self.errorMsg = ev.message; }
                }catch(e){}
              });
@@ -33,53 +33,78 @@ MOCK_CATALOG = [


 def _make_server(reject_non_chat: bool = True, accept: set | None = None,
-                 fail_all_chat: bool = False):
+                 fail_all_chat: bool = False, catalog: list | None = None,
+                 chat_status: dict | None = None):
    """HTTP server mimicking an OpenAI-compatible /models + /chat/completions.

    With ``reject_non_chat`` the chat endpoint returns 404 for any model whose
    id contains embed/rerank/flux, exactly like NVIDIA's public API.
    With ``fail_all_chat`` every chat call returns 404 (probe failure case).
    ``accept`` is an optional allow-list; when set, only those models answer 200.
+    ``catalog`` overrides the GET /models listing.
+    ``chat_status`` is an optional {model → status_code} map that takes precedence.
    """
+    catalog = catalog or MOCK_CATALOG
    accept_models = accept if accept is not None else set(MOCK_CATALOG[:2])
+    status_map = chat_status
+
+    def lookup_status(model: str) -> int | None:
+        if callable(status_map):
+            v = status_map(model)
+            return v if isinstance(v, int) else None
+        if isinstance(status_map, dict):
+            return status_map.get(model)
+        return None

    class Handler(BaseHTTPRequestHandler):
-        def log_message(self, *args):  # silence test noise
-            pass
+            def log_message(self, *args):  # silence test noise
+                pass

-        def do_GET(self):
-            if self.path.endswith("/models"):
-                body = json.dumps({"data": [{"id": m} for m in MOCK_CATALOG]}).encode()
-                self.send_response(200)
-                self.send_header("Content-Type", "application/json")
-                self.send_header("Content-Length", str(len(body)))
-                self.end_headers()
-                self.wfile.write(body)
-            else:
-                self.send_response(404)
-                self.send_header("Content-Length", "0")
-                self.end_headers()
+            def do_GET(self):
+                if self.path.endswith("/models"):
+                    body = json.dumps({"data": [{"id": m} for m in catalog]}).encode()
+                    self.send_response(200)
+                    self.send_header("Content-Type", "application/json")
+                    self.send_header("Content-Length", str(len(body)))
+                    self.end_headers()
+                    self.wfile.write(body)
+                else:
+                    self.send_response(404)
+                    self.send_header("Content-Length", "0")
+                    self.end_headers()

-        def do_POST(self):
-            if self.path.endswith("/chat/completions"):
-                n = int(self.headers.get("Content-Length", 0) or 0)
-                try:
-                    req = json.loads(self.rfile.read(n) or b"{}")
-                except ValueError:
-                    req = {}
-                model = req.get("model", "")
-                works = model in accept_models
-                if reject_non_chat and any(x in model for x in ("embed", "rerank", "flux")):
-                    works = False
-                if fail_all_chat:
-                    works = False
-                self.send_response(200 if works else 404)
-                self.send_header("Content-Length", "0")
-                self.end_headers()
-            else:
-                self.send_response(404)
-                self.send_header("Content-Length", "0")
-                self.end_headers()
+            def do_POST(self):
+                if self.path.endswith("/chat/completions"):
+                    n = int(self.headers.get("Content-Length", 0) or 0)
+                    try:
+                        req = json.loads(self.rfile.read(n) or b"{}")
+                    except ValueError:
+                        req = {}
+                    model = req.get("model", "")
+                    status = lookup_status(model)
+                    if status is None:
+                        works = model in accept_models
+                        if reject_non_chat and any(x in model for x in ("embed", "rerank", "flux")):
+                            works = False
+                        if fail_all_chat:
+                            works = False
+                        status = 200 if works else 404
+                    payload = b""
+                    if status == 200:
+                        payload = json.dumps({
+                            "choices": [{"index": 0, "message": {"role": "assistant", "content": "PONG"}}],
+                            "usage": {"total_tokens": 5},
+                        }).encode()
+                    self.send_response(status)
+                    self.send_header("Content-Type", "application/json")
+                    self.send_header("Content-Length", str(len(payload)))
+                    self.end_headers()
+                    if payload:
+                        self.wfile.write(payload)
+                else:
+                    self.send_response(404)
+                    self.send_header("Content-Length", "0")
+                    self.end_headers()

    server = ThreadingHTTPServer(("127.0.0.1", 0), Handler)
    threading.Thread(target=server.serve_forever, daemon=True).start()
@@ -135,5 +160,44 @@ def test_other_providers_keep_full_list(mock_server):
    assert models == MOCK_CATALOG


+def test_validation_keeps_429_but_drops_410():
+    # A model answering 429 is rate-limited but routed → kept.
+    # A model answering 410 (end of life) → definitively unusable → dropped.
+    def status_for(model):
+        return {"nvidia/good-chat": 200,
+                "nvidia/rate-limited-chat": 429,
+                "nvidia/retired-model": 410}.get(model, 404)
+
+    srv = _make_server(chat_status=status_for, catalog=["nvidia/good-chat",
+                       "nvidia/rate-limited-chat", "nvidia/retired-model"])
+    try:
+        base = f"http://127.0.0.1:{srv.server_port}/v1"
+        models = asyncio.run(fetch_provider_models(
+            "nvidia", api_key="nvapi-test", api_base=base, timeout=2))
+    finally:
+        srv.shutdown()
+    assert set(models) == {"nvidia/good-chat", "nvidia/rate-limited-chat"}
+
+
+def test_runtime_fallback_on_model_gone():
+    """LLMClient retries once with the provider default when the chosen model
+    answers 404/410, and the response carries a `notice` (visible in the UI)."""
+    from app.services.llm_client import LLMClient
+
+    def status_for(model):
+        return {"gone-or-deprecated": 410, "gpt-4o": 200}.get(model, 404)
+
+    srv = _make_server(chat_status=status_for)
+    try:
+        base = f"http://127.0.0.1:{srv.server_port}/v1"
+        llm = LLMClient(provider="openai", api_key="sk-test", api_base=base)
+        resp = asyncio.run(llm._http_complete(
+            [{"role": "user", "content": "ping"}], "gone-or-deprecated", None))
+    finally:
+        srv.shutdown()
+    assert resp.model == "gpt-4o"
+    assert "gpt-4o" in resp.notice
+
+
 def test_offline_returns_empty_list():
    assert asyncio.run(fetch_provider_models("offline", api_base="", timeout=2)) == []
@@ -1 +1 @@
 .1.8
 .1.9