Improve provider selection UX and fix GigaChat token refresh

bot/app.py

@@ -14,6 +14,16 @@ from telegram_api import TelegramAPI
 
 STATE_FILE = Path(__file__).resolve().parent.parent / ".new-qwen" / "telegram-state.json"
 TYPING_INTERVAL_SECONDS = 4.0
+BOT_COMMANDS = [
+    {"command": "help", "description": "Список команд"},
+    {"command": "status", "description": "Статус сервера и чата"},
+    {"command": "auth", "description": "Запустить Qwen OAuth"},
+    {"command": "provider", "description": "Выбрать провайдера"},
+    {"command": "model", "description": "Выбрать модель"},
+    {"command": "session", "description": "Показать сессию"},
+    {"command": "clear", "description": "Очистить контекст"},
+    {"command": "cancel", "description": "Отменить активный job"},
+]
 
 
 def load_state() -> dict[str, Any]:
@@ -26,6 +36,7 @@ def load_state() -> dict[str, Any]:
             "chat_active_jobs": {},
             "chat_queues": {},
             "pending_approvals": {},
+            "chat_preferences": {},
         }
     state = json.loads(STATE_FILE.read_text(encoding="utf-8"))
     state.setdefault("sessions", {})
@@ -34,6 +45,7 @@ def load_state() -> dict[str, Any]:
     state.setdefault("chat_active_jobs", {})
     state.setdefault("chat_queues", {})
     state.setdefault("pending_approvals", {})
+    state.setdefault("chat_preferences", {})
     return state
 
 
@@ -66,6 +78,49 @@ def send_text_chunks(api: TelegramAPI, chat_id: int, text: str) -> None:
         api.send_message(chat_id, normalized[start : start + chunk_size])
 
 
+def ensure_bot_commands(api: TelegramAPI, state: dict[str, Any]) -> None:
+    api.set_my_commands(BOT_COMMANDS)
+
+
+def get_provider_catalog(config: BotConfig) -> dict[str, Any]:
+    return get_json(f"{config.server_url}/api/v1/provider-catalog")
+
+
+def get_chat_preferences(state: dict[str, Any], chat_id: int) -> dict[str, Any]:
+    prefs = state.setdefault("chat_preferences", {}).setdefault(str(chat_id), {})
+    prefs.setdefault("provider", None)
+    prefs.setdefault("model", None)
+    return prefs
+
+
+def get_selected_provider_and_model(
+    state: dict[str, Any],
+    chat_id: int,
+    catalog: dict[str, Any],
+) -> tuple[str | None, str | None]:
+    prefs = get_chat_preferences(state, chat_id)
+    provider = prefs.get("provider") or catalog.get("default_provider")
+    model = prefs.get("model")
+    providers = {item.get("name"): item for item in catalog.get("providers", [])}
+    provider_info = providers.get(provider or "")
+    available_models = provider_info.get("models", []) if provider_info else []
+    if not model and available_models:
+        model = available_models[0].get("id")
+    return provider, model
+
+
+def format_status_text(job_state: dict[str, Any], text: str) -> str:
+    del job_state
+    return (text or "Пустой ответ.")[:4000]
+
+
+def format_final_signature(job_state: dict[str, Any]) -> str:
+    model = job_state.get("model") if isinstance(job_state.get("model"), str) else None
+    if not model:
+        return ""
+    return f"\n\n<i>{html.escape(model)}</i>"
+
+
 def render_markdownish_html(text: str) -> str:
     normalized = (text or "Пустой ответ.").replace("\r\n", "\n")
     fence_pattern = re.compile(r"```(?:[^\n`]*)\n(.*?)```", re.DOTALL)
@@ -91,7 +146,7 @@ def update_status_message(
     job_state: dict[str, Any],
     text: str,
 ) -> None:
-    normalized = (text or "Пустой ответ.")[:4000]
+    normalized = format_status_text(job_state, text)
     chat_id = int(job_state["chat_id"])
     message_id = int(job_state.get("status_message_id") or 0)
     if message_id:
@@ -100,6 +155,7 @@ def update_status_message(
         message_id = api.send_message(chat_id, normalized)
         job_state["status_message_id"] = message_id
     job_state["status_message_text"] = normalized
+    job_state["status_body_text"] = text or "Пустой ответ."
 
 
 def set_final_message(
@@ -109,8 +165,9 @@ def set_final_message(
 ) -> None:
     normalized = text or "Пустой ответ."
     chunk_size = 3800
+    signature = format_final_signature(job_state)
     first_chunk = normalized[:chunk_size]
-    rendered = render_markdownish_html(first_chunk)
+    rendered = render_markdownish_html(first_chunk) + signature
     chat_id = int(job_state["chat_id"])
     message_id = int(job_state.get("status_message_id") or 0)
     if message_id:
@@ -145,30 +202,38 @@ def keep_job_typing(
 def summarize_event(event: dict[str, Any]) -> str | None:
     event_type = event.get("type")
     if event_type == "job_status":
-        return event.get("message")
+        message = str(event.get("message") or "").strip()
+        if message == "Запрос принят сервером":
+            return "Смотрю, что можно сделать."
+        if message == "Ответ готов":
+            return "Формулирую ответ."
+        return message or None
     if event_type == "model_request":
-        provider = event.get("provider")
-        model = event.get("model")
-        if provider and model:
-            return f"Думаю над ответом через {provider}/{model}"
-        if provider:
-            return f"Думаю над ответом через {provider}"
-        return "Думаю над ответом"
+        return "Думаю над ответом."
     if event_type == "tool_call":
-        return f"Вызываю инструмент: {event.get('name')}"
+        tool_name = str(event.get("name") or "")
+        if tool_name in {"read_file", "list_files", "glob_search", "grep_text", "stat_path"}:
+            return "Просматриваю файлы и контекст."
+        if tool_name in {"git_status", "git_diff"}:
+            return "Проверяю изменения в проекте."
+        if tool_name in {"exec_command"}:
+            return "Проверяю окружение и команды."
+        if tool_name in {"web_search"}:
+            return "Ищу нужную информацию."
+        return "Проверяю детали."
     if event_type == "tool_result":
         result = event.get("result", {})
         if isinstance(result, dict) and "error" in result:
-            return f"Инструмент {event.get('name')} завершился с ошибкой"
-        return f"Инструмент {event.get('name')} завершён"
+            return "Наткнулся на проблему, перепроверяю."
+        return "Собрал нужные данные."
     if event_type == "error":
         return f"Ошибка: {event.get('message')}"
     if event_type == "approval_result":
         status = event.get("status")
         tool_name = event.get("tool_name")
         if status == "approved":
-            return f"Подтверждение получено для {tool_name}"
-        return f"Подтверждение отклонено для {tool_name}"
+            return f"Получил подтверждение для {tool_name}."
+        return f"Подтверждение для {tool_name} отклонено."
     return None
 
 
@@ -183,6 +248,83 @@ def format_approval_keyboard(approval_id: str) -> dict[str, Any]:
     }
 
 
+def format_provider_keyboard(catalog: dict[str, Any]) -> dict[str, Any]:
+    rows: list[list[dict[str, str]]] = []
+    row: list[dict[str, str]] = []
+    for item in catalog.get("providers", []):
+        name = str(item.get("name") or "")
+        if not name:
+            continue
+        row.append({"text": name, "callback_data": f"provider:{name}"})
+        if len(row) == 2:
+            rows.append(row)
+            row = []
+    if row:
+        rows.append(row)
+    return {"inline_keyboard": rows}
+
+
+def format_model_keyboard(provider: str, models: list[dict[str, str]]) -> dict[str, Any]:
+    rows = [
+        [{"text": str(item.get("name") or item.get("id") or "-"), "callback_data": f"model:{provider}:{item.get('id')}"}]
+        for item in models
+        if item.get("id")
+    ]
+    return {"inline_keyboard": rows}
+
+
+def set_chat_provider(
+    state: dict[str, Any],
+    chat_id: int,
+    provider: str | None,
+    *,
+    reset_model: bool = True,
+) -> dict[str, Any]:
+    prefs = get_chat_preferences(state, chat_id)
+    prefs["provider"] = provider
+    if reset_model:
+        prefs["model"] = None
+    return prefs
+
+
+def set_chat_model(state: dict[str, Any], chat_id: int, model: str | None) -> dict[str, Any]:
+    prefs = get_chat_preferences(state, chat_id)
+    prefs["model"] = model
+    return prefs
+
+
+def format_provider_selection(
+    catalog: dict[str, Any],
+    selected_provider: str | None,
+) -> str:
+    lines = ["Выбери провайдера."]
+    for item in catalog.get("providers", []):
+        name = str(item.get("name") or "")
+        if not name:
+            continue
+        marker = "•" if name == selected_provider else " "
+        availability = "ready" if item.get("available") else f"unavailable: {item.get('reason') or 'unknown'}"
+        lines.append(f"{marker} {name} ({availability})")
+    return "\n".join(lines)
+
+
+def format_model_selection(
+    provider: str,
+    models: list[dict[str, str]],
+    selected_model: str | None,
+) -> str:
+    lines = [f"Выбери модель для {provider}."]
+    for item in models:
+        model_id = str(item.get("id") or "")
+        if not model_id:
+            continue
+        marker = "•" if model_id == selected_model else " "
+        description = str(item.get("description") or "").strip()
+        suffix = f" ({description})" if description else ""
+        lines.append(f"{marker} {model_id}{suffix}")
+    return "\n".join(lines)
+
+
 def format_approval_request(event: dict[str, Any]) -> str:
     return (
         "Нужно подтверждение инструмента.\n"
@@ -290,7 +432,9 @@ def start_chat_job(
     delayed: bool = False,
 ) -> None:
     session_id = state.setdefault("sessions", {}).get(session_key)
-    prefix = "Обрабатываю отложенный запрос..." if delayed else "Обрабатываю запрос..."
+    catalog = get_provider_catalog(config)
+    provider, model = get_selected_provider_and_model(state, chat_id, catalog)
+    prefix = "Возвращаюсь к вашему сообщению." if delayed else "Смотрю, что можно сделать."
     status_message_id = api.send_message(chat_id, prefix)
     start_result = post_json(
         f"{config.server_url}/api/v1/chat/start",
@@ -298,6 +442,8 @@ def start_chat_job(
             "session_id": session_id,
             "user_id": user_id,
             "message": text,
+            "provider": provider,
+            "model": model,
         },
     )
     state["sessions"][session_key] = start_result["session_id"]
@@ -310,6 +456,12 @@ def start_chat_job(
         "seen_seq": 0,
         "status_message_id": status_message_id,
         "status_message_text": prefix,
+        "status_body_text": prefix,
+        "requested_provider": provider,
+        "requested_model": model,
+        "provider": start_result.get("provider") or provider,
+        "model": start_result.get("model") or model,
+        "fallback_notified": False,
         "last_typing_at": 0.0,
     }
     state.setdefault("chat_active_jobs", {})[str(chat_id)] = start_result["job_id"]
@@ -501,8 +653,25 @@ def process_active_jobs(
                     reply_markup=format_approval_keyboard(str(event["approval_id"])),
                 )
                 continue
+            if event.get("type") == "model_request":
+                selection_reason = str(event.get("selection_reason") or "")
+                if (
+                    "fell back to" in selection_reason
+                    and not job_state.get("fallback_notified")
+                ):
+                    requested_provider = job_state.get("requested_provider") or "unknown"
+                    api.send_message(
+                        int(job_state["chat_id"]),
+                        "Выбранный провайдер не ответил, поэтому продолжаю через запасной вариант.\n"
+                        f"Изначально был выбран: {requested_provider}",
+                    )
+                    job_state["fallback_notified"] = True
+                if event.get("provider"):
+                    job_state["provider"] = event.get("provider")
+                if event.get("model"):
+                    job_state["model"] = event.get("model")
             summary = summarize_event(event)
-            if summary and summary != job_state.get("status_message_text"):
+            if summary and summary != job_state.get("status_body_text"):
                 update_status_message(api, job_state, summary)
 
         status = poll_result.get("status")
@@ -590,7 +759,7 @@ def handle_message(api: TelegramAPI, config: BotConfig, state: dict[str, Any], m
     if text == "/start":
         api.send_message(
             chat_id,
-            "new-qwen bot готов.\nКоманды: /help, /auth, /status, /session, /cancel, /clear, /approve, /reject.",
+            "new-qwen bot готов.\nКоманды доступны через Telegram menu: /help, /auth, /status, /provider, /model, /session, /cancel, /clear.",
         )
         return
 
@@ -601,6 +770,8 @@ def handle_message(api: TelegramAPI, config: BotConfig, state: dict[str, Any], m
             "/auth - начать Qwen OAuth\n"
             "/auth_check [flow_id] - проверить авторизацию\n"
             "/status - статус OAuth и сервера\n"
+            "/provider [name] - выбрать провайдера\n"
+            "/model [id] - выбрать модель\n"
             "/session - показать текущую сессию\n"
             "/cancel - отменить активный запрос и очистить очередь\n"
             "/approve [approval_id] - подтвердить инструмент\n"
@@ -652,10 +823,16 @@ def handle_message(api: TelegramAPI, config: BotConfig, state: dict[str, Any], m
 
     if text == "/status":
         status = get_json(f"{config.server_url}/api/v1/auth/status")
-        models_status = get_json(f"{config.server_url}/api/v1/models")
+        catalog = get_provider_catalog(config)
         queue_size = len(state.setdefault("chat_queues", {}).get(str(chat_id), []))
         active_job = state.setdefault("chat_active_jobs", {}).get(str(chat_id))
-        models_info = "\n".join([f"  - {m['id']}: {m['name']} ({m['description']})" for m in models_status.get("models", [])]) or "  Нет доступных моделей"
+        provider, model = get_selected_provider_and_model(state, chat_id, catalog)
+        providers_info = "\n".join(
+            [
+                f"  - {item['name']}: model={item.get('model')}, available={item.get('available')}"
+                for item in catalog.get("providers", [])
+            ]
+        ) or "  Нет доступных провайдеров"
         send_text_chunks(
             api,
             chat_id,
@@ -665,15 +842,61 @@ def handle_message(api: TelegramAPI, config: BotConfig, state: dict[str, Any], m
             f"available_providers: {', '.join(status.get('available_providers') or []) or '-'}\n"
             f"default_provider: {status.get('default_provider')}\n"
             f"fallback_providers: {', '.join(status.get('fallback_providers') or []) or '-'}\n"
+            f"selected_provider: {provider}\n"
+            f"selected_model: {model}\n"
             f"resource_url: {status.get('resource_url')}\n"
             f"expires_at: {status.get('expires_at')}\n"
             f"tool_policy: {status.get('tool_policy')}\n"
-            f"Доступные Qwen OAuth модели:\n{models_info}\n"
+            f"Провайдеры и модели:\n{providers_info}\n"
             f"active_job: {active_job}\n"
             f"queued_messages: {queue_size}",
         )
         return
 
+    if text.startswith("/provider"):
+        catalog = get_provider_catalog(config)
+        parts = text.split(maxsplit=1)
+        if len(parts) == 1:
+            provider, _ = get_selected_provider_and_model(state, chat_id, catalog)
+            api.send_message(
+                chat_id,
+                format_provider_selection(catalog, provider),
+                reply_markup=format_provider_keyboard(catalog),
+            )
+            return
+        provider_name = parts[1].strip()
+        provider_names = {str(item.get("name") or "") for item in catalog.get("providers", [])}
+        if provider_name not in provider_names:
+            api.send_message(chat_id, f"Неизвестный provider: {provider_name}")
+            return
+        set_chat_provider(state, chat_id, provider_name, reset_model=True)
+        _, model = get_selected_provider_and_model(state, chat_id, catalog)
+        api.send_message(chat_id, f"Выбран provider: {provider_name}\nmodel: {model or '-'}")
+        return
+
+    if text.startswith("/model"):
+        catalog = get_provider_catalog(config)
+        provider, selected_model = get_selected_provider_and_model(state, chat_id, catalog)
+        provider_map = {str(item.get("name") or ""): item for item in catalog.get("providers", [])}
+        provider_info = provider_map.get(provider or "")
+        models = provider_info.get("models", []) if provider_info else []
+        parts = text.split(maxsplit=1)
+        if len(parts) == 1:
+            api.send_message(
+                chat_id,
+                format_model_selection(provider or "-", models, selected_model),
+                reply_markup=format_model_keyboard(provider or "-", models),
+            )
+            return
+        model_id = parts[1].strip()
+        valid_model_ids = {str(item.get("id") or "") for item in models}
+        if model_id not in valid_model_ids:
+            api.send_message(chat_id, f"Неизвестная модель для {provider}: {model_id}")
+            return
+        set_chat_model(state, chat_id, model_id)
+        api.send_message(chat_id, f"Выбрана модель: {model_id}\nprovider: {provider}")
+        return
+
     if text == "/cancel":
         canceled = cancel_chat_work(
             api,
@@ -750,6 +973,44 @@ def handle_callback_query(
 
     action, approval_id = data.split(":", 1)
     if action not in {"approve", "reject"}:
+        if action == "provider":
+            catalog = get_provider_catalog(config)
+            provider_names = {str(item.get("name") or "") for item in catalog.get("providers", [])}
+            if approval_id not in provider_names:
+                api.answer_callback_query(callback_query_id, "Неизвестный провайдер")
+                return
+            set_chat_provider(state, chat_id, approval_id, reset_model=True)
+            provider, model = get_selected_provider_and_model(state, chat_id, catalog)
+            api.edit_message_text(
+                chat_id,
+                message_id,
+                f"Выбран provider: {provider}\nmodel: {model or '-'}",
+                reply_markup={"inline_keyboard": []},
+            )
+            api.answer_callback_query(callback_query_id, "Провайдер переключен")
+            return
+        if action == "model":
+            provider, model_id = approval_id.split(":", 1) if ":" in approval_id else ("", "")
+            catalog = get_provider_catalog(config)
+            provider_map = {str(item.get("name") or ""): item for item in catalog.get("providers", [])}
+            provider_info = provider_map.get(provider)
+            model_ids = {
+                str(item.get("id") or "")
+                for item in (provider_info.get("models", []) if provider_info else [])
+            }
+            if model_id not in model_ids:
+                api.answer_callback_query(callback_query_id, "Неизвестная модель")
+                return
+            set_chat_provider(state, chat_id, provider, reset_model=False)
+            set_chat_model(state, chat_id, model_id)
+            api.edit_message_text(
+                chat_id,
+                message_id,
+                f"Выбрана модель: {model_id}\nprovider: {provider}",
+                reply_markup={"inline_keyboard": []},
+            )
+            api.answer_callback_query(callback_query_id, "Модель переключена")
+            return
         api.answer_callback_query(callback_query_id, "Неизвестное действие")
         return
 
@@ -782,6 +1043,8 @@ def main() -> None:
     config = BotConfig.load()
     api = TelegramAPI(config.token, proxy_url=config.proxy_url)
     state = load_state()
+    ensure_bot_commands(api, state)
+    save_state(state)
     print("new-qwen bot polling started")
     while True:
         try:

bot/telegram_api.py

@@ -119,3 +119,6 @@ class TelegramAPI:
         if text:
             payload["text"] = text
         self._post("answerCallbackQuery", payload)
+
+    def set_my_commands(self, commands: list[dict[str, str]]) -> None:
+        self._post("setMyCommands", {"commands": commands})

serv/app.py

@@ -139,6 +139,13 @@ class AppState:
             "pending_approvals": len(self.approvals.list_pending()),
         }
 
+    def provider_catalog(self) -> dict[str, Any]:
+        return {
+            "default_provider": self.config.default_provider,
+            "default_model": self.config.model,
+            "providers": self.providers.statuses(),
+        }
+
 
 class RequestHandler(BaseHTTPRequestHandler):
     server_version = "new-qwen-serv/0.1"
@@ -180,6 +187,9 @@ class RequestHandler(BaseHTTPRequestHandler):
             models = self.app.oauth.get_available_models()
             self._send(HTTPStatus.OK, {"models": models})
             return
+        if self.path == "/api/v1/provider-catalog":
+            self._send(HTTPStatus.OK, self.app.provider_catalog())
+            return
         self._send(HTTPStatus.NOT_FOUND, {"error": "Not found"})
 
     def _run_chat_job(
@@ -189,6 +199,7 @@ class RequestHandler(BaseHTTPRequestHandler):
         user_id: str,
         message: str,
         preferred_provider: str | None = None,
+        preferred_model: str | None = None,
     ) -> None:
         try:
             if self.app.jobs.is_cancel_requested(job_id):
@@ -217,6 +228,7 @@ class RequestHandler(BaseHTTPRequestHandler):
                 ),
                 is_cancelled=lambda: self.app.jobs.is_cancel_requested(job_id),
                 preferred_provider=preferred_provider,
+                preferred_model=preferred_model,
             )
             if self.app.jobs.is_cancel_requested(job_id):
                 reason = "Job canceled by operator"
@@ -338,12 +350,14 @@ class RequestHandler(BaseHTTPRequestHandler):
                 user_id = str(body.get("user_id") or "anonymous")
                 message = body["message"]
                 preferred_provider = body.get("provider")
+                preferred_model = body.get("model")
                 session = self.app.sessions.load(session_id)
                 history = session.get("messages", [])
                 result = self.app.agent.run(
                     history,
                     message,
                     preferred_provider=preferred_provider,
+                    preferred_model=preferred_model,
                 )
                 persisted_messages = result["messages"][1:]
                 self.app.sessions.save(
@@ -375,10 +389,11 @@ class RequestHandler(BaseHTTPRequestHandler):
                 user_id = str(body.get("user_id") or "anonymous")
                 message = body["message"]
                 preferred_provider = body.get("provider")
+                preferred_model = body.get("model")
                 job = self.app.jobs.create(session_id, user_id, message)
                 thread = threading.Thread(
                     target=self._run_chat_job,
-                    args=(job["job_id"], session_id, user_id, message, preferred_provider),
+                    args=(job["job_id"], session_id, user_id, message, preferred_provider, preferred_model),
                     daemon=True,
                 )
                 thread.start()
@@ -389,6 +404,7 @@ class RequestHandler(BaseHTTPRequestHandler):
                         "session_id": session_id,
                         "status": "queued",
                         "provider": preferred_provider or self.app.config.default_provider,
+                        "model": preferred_model or self.app.config.model,
                     },
                 )
                 return

serv/gigachat.py

@@ -55,6 +55,14 @@ class GigaChatAuthManager:
             encoding="utf-8",
         )
 
+    @staticmethod
+    def _normalize_expires_at(value: Any) -> int:
+        expires_at = int(value or 0)
+        # Sber can return unix time in milliseconds.
+        if expires_at > 10_000_000_000:
+            return expires_at // 1000
+        return expires_at
+
     def fetch_token(self) -> dict[str, Any]:
         data = parse.urlencode({"scope": self.config.gigachat_scope}).encode("utf-8")
         req = request.Request(
@@ -76,7 +84,7 @@ class GigaChatAuthManager:
             raise GigaChatError(f"GigaChat token request failed with HTTP {exc.code}: {body}") from exc
         token = {
             "access_token": payload["access_token"],
-            "expires_at": int(payload["expires_at"]),
+            "expires_at": self._normalize_expires_at(payload.get("expires_at")),
         }
         self.save_token(token)
         return token
@@ -86,7 +94,8 @@ class GigaChatAuthManager:
             raise GigaChatError("GigaChat auth key is not configured")
         token = self.load_token()
         now = int(time.time())
-        if token and int(token.get("expires_at", 0)) - now > 30:
+        expires_at = self._normalize_expires_at(token.get("expires_at", 0)) if token else 0
+        if token and expires_at - now > 30:
             return str(token["access_token"])
         refreshed = self.fetch_token()
         return str(refreshed["access_token"])

serv/llm.py

@@ -30,6 +30,7 @@ class QwenAgent:
         approval_callback: Callable[[str, dict[str, Any]], dict[str, Any]] | None = None,
         is_cancelled: Callable[[], bool] | None = None,
         preferred_provider: str | None = None,
+        preferred_model: str | None = None,
     ) -> dict[str, Any]:
         emit = on_event or (lambda _event: None)
         cancel_check = is_cancelled or (lambda: False)
@@ -54,6 +55,7 @@ class QwenAgent:
                     tools=self.tools.schemas(),
                     tool_choice="auto",
                     preferred_provider=preferred_provider,
+                    preferred_model=preferred_model,
                     require_tools=True,
                 )
             )

serv/model_router.py

@@ -1,6 +1,7 @@
 from __future__ import annotations
 
 import json
+from dataclasses import replace
 from dataclasses import dataclass, field
 from typing import Any
 import uuid
@@ -28,6 +29,7 @@ class CompletionRequest:
     tools: list[dict[str, Any]]
     tool_choice: str = "auto"
     preferred_provider: str | None = None
+    preferred_model: str | None = None
     require_tools: bool = True
 
 
@@ -53,9 +55,14 @@ class BaseModelProvider:
     def unavailable_reason(self) -> str | None:
         raise NotImplementedError
 
-    def model_name(self) -> str:
+    def model_name(self, preferred_model: str | None = None) -> str:
+        del preferred_model
         raise NotImplementedError
 
+    def list_models(self) -> list[dict[str, str]]:
+        model = self.model_name()
+        return [{"id": model, "name": model, "description": ""}]
+
     def complete(self, completion_request: CompletionRequest) -> dict[str, Any]:
         raise NotImplementedError
 
@@ -83,7 +90,8 @@ class UnavailableModelProvider(BaseModelProvider):
     def unavailable_reason(self) -> str | None:
         return self._reason
 
-    def model_name(self) -> str:
+    def model_name(self, preferred_model: str | None = None) -> str:
+        del preferred_model
         return self._model_name
 
     def complete(self, completion_request: CompletionRequest) -> dict[str, Any]:
@@ -114,8 +122,15 @@ class QwenModelProvider(BaseModelProvider):
             return None
         return "Qwen OAuth is not configured"
 
-    def model_name(self) -> str:
-        return self._model_name
+    def model_name(self, preferred_model: str | None = None) -> str:
+        if preferred_model and preferred_model not in QWEN_OAUTH_ALLOWED_MODELS:
+            raise ModelProviderError(
+                f"Provider {self.name} does not support model '{preferred_model}'"
+            )
+        return self.oauth.get_model_name_for_id(preferred_model or self._model_id)
+
+    def list_models(self) -> list[dict[str, str]]:
+        return self.oauth.get_available_models()
 
     @staticmethod
     def _normalize_content(value: Any) -> list[dict[str, str]]:
@@ -140,8 +155,9 @@ class QwenModelProvider(BaseModelProvider):
     def complete(self, completion_request: CompletionRequest) -> dict[str, Any]:
         creds = self.oauth.get_valid_credentials()
         base_url = self.oauth.get_openai_base_url(creds)
+        model_name = self.model_name(completion_request.preferred_model)
         payload = {
-            "model": self.model_name(),
+            "model": model_name,
             "messages": self._normalize_messages(completion_request.messages),
             "max_tokens": 8000,
             "metadata": {
@@ -203,8 +219,18 @@ class GigaChatModelProvider(BaseModelProvider):
             return None
         return "GigaChat auth key is not configured"
 
-    def model_name(self) -> str:
-        return self.config.gigachat_model
+    def model_name(self, preferred_model: str | None = None) -> str:
+        return (preferred_model or self.config.gigachat_model).strip() or self.config.gigachat_model
+
+    def list_models(self) -> list[dict[str, str]]:
+        model = self.config.gigachat_model
+        return [
+            {
+                "id": model,
+                "name": model,
+                "description": "Configured GigaChat model",
+            }
+        ]
 
     @staticmethod
     def _convert_tool_schema(tool: dict[str, Any]) -> dict[str, Any]:
@@ -294,7 +320,7 @@ class GigaChatModelProvider(BaseModelProvider):
             raise ModelProviderError(str(exc)) from exc
         api_base = self.config.gigachat_api_base_url.rstrip("/")
         payload: dict[str, Any] = {
-            "model": self.model_name(),
+            "model": self.model_name(completion_request.preferred_model),
             "messages": self._convert_messages(completion_request.messages),
         }
         if completion_request.tools:
@@ -340,6 +366,7 @@ class ProviderRegistry:
             {
                 "name": provider.name,
                 "model": provider.model_name(),
+                "models": provider.list_models(),
                 "available": provider.is_available(),
                 "reason": provider.unavailable_reason(),
                 "capabilities": {
@@ -400,13 +427,16 @@ class ModelRouter:
             elif name == self.config.default_provider:
                 selection_reason = "selected default provider"
             try:
-                payload = provider.complete(completion_request)
+                provider_request = completion_request
+                if completion_request.preferred_provider and name != completion_request.preferred_provider:
+                    provider_request = replace(completion_request, preferred_model=None)
+                payload = provider.complete(provider_request)
             except ModelProviderError as exc:
                 reasons.append(f"{name}: request failed: {exc}")
                 continue
             return CompletionResponse(
                 provider_name=name,
-                model_name=provider.model_name(),
+                model_name=provider.model_name(provider_request.preferred_model),
                 payload=payload,
                 selection_reason=selection_reason,
                 attempted=attempted,