ccs-llmconnector 1.1.0__py3-none-any.whl → 1.1.1__py3-none-any.whl

llmconnector/gemini_client.py

@@ -2,54 +2,54 @@
 
 from __future__ import annotations
 
-import base64
-import mimetypes
-from pathlib import Path
-import logging
-from typing import Optional, Sequence
-from urllib.request import urlopen
-
-from google import genai
-from google.genai import types
-
-from .types import ImageInput, MessageSequence, normalize_messages
-from .utils import clamp_retries, run_sync_in_thread, run_with_retries
-
-logger = logging.getLogger(__name__)
+import base64
+import mimetypes
+from pathlib import Path
+import logging
+from typing import Optional, Sequence
+from urllib.request import urlopen
+
+from google import genai
+from google.genai import types
+
+from .types import ImageInput, MessageSequence, normalize_messages
+from .utils import clamp_retries, run_sync_in_thread, run_with_retries
+
+logger = logging.getLogger(__name__)
 
 
 class GeminiClient:
     """Convenience wrapper around the Google Gemini SDK."""
 
-    def generate_response(
-        self,
-        *,
-        api_key: str,
-        prompt: Optional[str] = None,
-        model: str,
-        max_tokens: int = 32000,
-        reasoning_effort: Optional[str] = None,
-        images: Optional[Sequence[ImageInput]] = None,
-        messages: Optional[MessageSequence] = None,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> str:
+    def generate_response(
+        self,
+        *,
+        api_key: str,
+        prompt: Optional[str] = None,
+        model: str,
+        max_tokens: int = 32000,
+        reasoning_effort: Optional[str] = None,
+        images: Optional[Sequence[ImageInput]] = None,
+        messages: Optional[MessageSequence] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> str:
         """Generate a response from the specified Gemini model.
 
         Args:
             api_key: API key used to authenticate with the Gemini API.
-            prompt: Natural-language instruction or query for the model.
-            model: Identifier of the Gemini model to target (for example, ``"gemini-2.5-flash"``).
-            max_tokens: Cap for tokens across the entire exchange, defaults to 32000.
-            reasoning_effort: Included for API parity; currently unused by the Gemini SDK.
-            images: Optional collection of image references (local paths, URLs, or data URLs).
-            messages: Optional list of chat-style messages (role/content).
-            request_id: Optional request identifier for tracing/logging.
-            timeout_s: Optional request timeout in seconds.
-            max_retries: Optional retry count for transient failures.
-            retry_backoff_s: Base delay (seconds) for exponential backoff between retries.
+            prompt: Natural-language instruction or query for the model.
+            model: Identifier of the Gemini model to target (for example, ``"gemini-2.5-flash"``).
+            max_tokens: Cap for tokens across the entire exchange, defaults to 32000.
+            reasoning_effort: Included for API parity; currently unused by the Gemini SDK.
+            images: Optional collection of image references (local paths, URLs, or data URLs).
+            messages: Optional list of chat-style messages (role/content).
+            request_id: Optional request identifier for tracing/logging.
+            timeout_s: Optional request timeout in seconds.
+            max_retries: Optional retry count for transient failures.
+            retry_backoff_s: Base delay (seconds) for exponential backoff between retries.
 
         Returns:
             The text output produced by the model.
@@ -58,179 +58,178 @@ class GeminiClient:
             ValueError: If required arguments are missing or the request payload is empty.
             URLError: If an image URL cannot be retrieved.
             google.genai.errors.APIError: If the underlying Gemini request fails.
-        """
-        if not api_key:
-            raise ValueError("api_key must be provided.")
-        if not prompt and not messages and not images:
-            raise ValueError("At least one of prompt, messages, or images must be provided.")
-        if not model:
-            raise ValueError("model must be provided.")
-
-        normalized_messages = normalize_messages(prompt=prompt, messages=messages)
-        contents: list[types.Content] = []
-        for message in normalized_messages:
-            parts: list[types.Part] = []
-            if message["content"]:
-                parts.append(types.Part.from_text(text=message["content"]))
-            contents.append(types.Content(role=message["role"], parts=parts))
-
-        if images:
-            image_parts = [self._to_image_part(image) for image in images]
-            target_index = next(
-                (
-                    index
-                    for index in range(len(contents) - 1, -1, -1)
-                    if contents[index].role == "user"
-                ),
-                None,
-            )
-            if target_index is None:
-                contents.append(types.Content(role="user", parts=image_parts))
-            else:
-                existing_parts = list(contents[target_index].parts or [])
-                existing_parts.extend(image_parts)
-                contents[target_index] = types.Content(
-                    role="user", parts=existing_parts
-                )
-
-        if not contents or not any(content.parts for content in contents):
-            raise ValueError("No content provided for response generation.")
-
-        config = types.GenerateContentConfig(max_output_tokens=max_tokens)
-        # reasoning_effort is accepted for compatibility but not currently applied because the
-        # Gemini SDK does not expose an equivalent configuration parameter.
-
-        retry_count = clamp_retries(max_retries)
-
-        def _build_client() -> genai.Client:
-            client_kwargs: dict[str, object] = {"api_key": api_key}
-            if timeout_s is not None:
-                http_options = getattr(types, "HttpOptions", None)
-                if http_options is not None:
-                    try:
-                        client_kwargs["http_options"] = http_options(timeout=timeout_s)
-                    except Exception:
-                        logger.debug("Gemini HttpOptions timeout not applied.", exc_info=True)
-            return genai.Client(**client_kwargs)
-
-        def _run_request() -> str:
-            client = _build_client()
-            try:
-                try:
-                    response = client.models.generate_content(
-                        model=model,
-                        contents=contents,
-                        config=config,
-                    )
-                except Exception as exc:
-                    logger.exception(
-                        "Gemini generate_content failed: %s request_id=%s",
-                        exc,
-                        request_id,
-                    )
-                    raise
-            finally:
-                closer = getattr(client, "close", None)
-                if callable(closer):
-                    try:
-                        closer()
-                    except Exception:
-                        pass
-
-            if response.text:
-                result_text = response.text
-                logger.info(
-                    "Gemini generate_content succeeded: model=%s images=%d text_len=%d request_id=%s",
-                    model,
-                    len(images or []),
-                    len(result_text or ""),
-                    request_id,
-                )
-                return result_text
-
-            candidate_texts: list[str] = []
-            for candidate in getattr(response, "candidates", []) or []:
-                content_obj = getattr(candidate, "content", None)
-                if not content_obj:
-                    continue
-                for part in getattr(content_obj, "parts", []) or []:
-                    text = getattr(part, "text", None)
-                    if text:
-                        candidate_texts.append(text)
-
-            if candidate_texts:
-                result_text = "\n".join(candidate_texts)
-                logger.info(
-                    "Gemini generate_content succeeded (candidates): model=%s images=%d text_len=%d request_id=%s",
-                    model,
-                    len(images or []),
-                    len(result_text or ""),
-                    request_id,
-                )
-                return result_text
-
-            # Treat successful calls without textual content as a successful, empty response
-            # rather than raising. This aligns with callers that handle empty outputs gracefully.
-            logger.info(
-                "Gemini generate_content succeeded with no text: model=%s images=%d request_id=%s",
-                model,
-                len(images or []),
-                request_id,
-            )
-            return ""
-
-        return run_with_retries(
-            func=_run_request,
-            max_retries=retry_count,
-            retry_backoff_s=retry_backoff_s,
-            request_id=request_id,
-        )
-
-    async def async_generate_response(
-        self,
-        *,
-        api_key: str,
-        prompt: Optional[str] = None,
-        model: str,
-        max_tokens: int = 32000,
-        reasoning_effort: Optional[str] = None,
-        images: Optional[Sequence[ImageInput]] = None,
-        messages: Optional[MessageSequence] = None,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> str:
-        return await run_sync_in_thread(
-            lambda: self.generate_response(
-                api_key=api_key,
-                prompt=prompt,
-                model=model,
-                max_tokens=max_tokens,
-                reasoning_effort=reasoning_effort,
-                images=images,
-                messages=messages,
-                request_id=request_id,
-                timeout_s=timeout_s,
-                max_retries=max_retries,
-                retry_backoff_s=retry_backoff_s,
-            )
-        )
-
-    def generate_image(
-        self,
-        *,
-        api_key: str,
-        prompt: str,
-        model: str,
-        image_size: Optional[str] = None,
-        aspect_ratio: Optional[str] = None,
-        image: Optional[ImageInput] = None,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> bytes:
+        """
+        if not api_key:
+            raise ValueError("api_key must be provided.")
+        if not prompt and not messages and not images:
+            raise ValueError("At least one of prompt, messages, or images must be provided.")
+        if not model:
+            raise ValueError("model must be provided.")
+
+        normalized_messages = normalize_messages(prompt=prompt, messages=messages)
+        contents: list[types.Content] = []
+        for message in normalized_messages:
+            parts: list[types.Part] = []
+            if message["content"]:
+                parts.append(types.Part.from_text(text=message["content"]))
+            contents.append(types.Content(role=message["role"], parts=parts))
+
+        if images:
+            image_parts = [self._to_image_part(image) for image in images]
+            target_index = next(
+                (
+                    index
+                    for index in range(len(contents) - 1, -1, -1)
+                    if contents[index].role == "user"
+                ),
+                None,
+            )
+            if target_index is None:
+                contents.append(types.Content(role="user", parts=image_parts))
+            else:
+                existing_parts = list(contents[target_index].parts or [])
+                existing_parts.extend(image_parts)
+                contents[target_index] = types.Content(
+                    role="user", parts=existing_parts
+                )
+
+        if not contents or not any(content.parts for content in contents):
+            raise ValueError("No content provided for response generation.")
+
+        config = types.GenerateContentConfig(max_output_tokens=max_tokens)
+        # reasoning_effort is accepted for compatibility but not currently applied because the
+        # Gemini SDK does not expose an equivalent configuration parameter.
+
+        retry_count = clamp_retries(max_retries)
+
+        def _build_client() -> genai.Client:
+            client_kwargs: dict[str, object] = {"api_key": api_key}
+            if timeout_s is not None:
+                # Gemini requires at least 10s timeout if set
+                effective_timeout = max(10.0, timeout_s)
+                if effective_timeout != timeout_s:
+                    logger.warning("Gemini timeout %ss is too short, clamping to %ss.", timeout_s, effective_timeout)
+                client_kwargs["http_options"] = types.HttpOptions(timeout=effective_timeout)
+            return genai.Client(**client_kwargs)
+
+        def _run_request() -> str:
+            client = _build_client()
+            try:
+                try:
+                    response = client.models.generate_content(
+                        model=model,
+                        contents=contents,
+                        config=config,
+                    )
+                except Exception as exc:
+                    logger.exception(
+                        "Gemini generate_content failed: %s request_id=%s",
+                        exc,
+                        request_id,
+                    )
+                    raise
+            finally:
+                closer = getattr(client, "close", None)
+                if callable(closer):
+                    try:
+                        closer()
+                    except Exception:
+                        pass
+
+            if response.text:
+                result_text = response.text
+                logger.info(
+                    "Gemini generate_content succeeded: model=%s images=%d text_len=%d request_id=%s",
+                    model,
+                    len(images or []),
+                    len(result_text or ""),
+                    request_id,
+                )
+                return result_text
+
+            candidate_texts: list[str] = []
+            for candidate in getattr(response, "candidates", []) or []:
+                content_obj = getattr(candidate, "content", None)
+                if not content_obj:
+                    continue
+                for part in getattr(content_obj, "parts", []) or []:
+                    text = getattr(part, "text", None)
+                    if text:
+                        candidate_texts.append(text)
+
+            if candidate_texts:
+                result_text = "\n".join(candidate_texts)
+                logger.info(
+                    "Gemini generate_content succeeded (candidates): model=%s images=%d text_len=%d request_id=%s",
+                    model,
+                    len(images or []),
+                    len(result_text or ""),
+                    request_id,
+                )
+                return result_text
+
+            # Treat successful calls without textual content as a successful, empty response
+            # rather than raising. This aligns with callers that handle empty outputs gracefully.
+            logger.info(
+                "Gemini generate_content succeeded with no text: model=%s images=%d request_id=%s",
+                model,
+                len(images or []),
+                request_id,
+            )
+            return ""
+
+        return run_with_retries(
+            func=_run_request,
+            max_retries=retry_count,
+            retry_backoff_s=retry_backoff_s,
+            request_id=request_id,
+        )
+
+    async def async_generate_response(
+        self,
+        *,
+        api_key: str,
+        prompt: Optional[str] = None,
+        model: str,
+        max_tokens: int = 32000,
+        reasoning_effort: Optional[str] = None,
+        images: Optional[Sequence[ImageInput]] = None,
+        messages: Optional[MessageSequence] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> str:
+        return await run_sync_in_thread(
+            lambda: self.generate_response(
+                api_key=api_key,
+                prompt=prompt,
+                model=model,
+                max_tokens=max_tokens,
+                reasoning_effort=reasoning_effort,
+                images=images,
+                messages=messages,
+                request_id=request_id,
+                timeout_s=timeout_s,
+                max_retries=max_retries,
+                retry_backoff_s=retry_backoff_s,
+            )
+        )
+
+    def generate_image(
+        self,
+        *,
+        api_key: str,
+        prompt: str,
+        model: str,
+        image_size: Optional[str] = None,
+        aspect_ratio: Optional[str] = None,
+        image: Optional[ImageInput] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> bytes:
         """Generate an image using Gemini 3 Pro Image.
 
         Args:
@@ -255,195 +254,193 @@ class GeminiClient:
         if not model:
             raise ValueError("model must be provided.")
 
-        config = types.GenerateContentConfig(
-            tools=[{"google_search": {}}],
-            image_config=types.ImageConfig(
-                image_size=image_size or "2K",
-                aspect_ratio=aspect_ratio,
-            ),
-        )
+        config = types.GenerateContentConfig(
+            tools=[{"google_search": {}}],
+            image_config=types.ImageConfig(
+                image_size=image_size or "2K",
+                aspect_ratio=aspect_ratio,
+            ),
+        )
 
         contents = [prompt]
         if image:
             contents.append(self._to_image_part(image))
 
-        retry_count = clamp_retries(max_retries)
-
-        def _build_client() -> genai.Client:
-            client_kwargs: dict[str, object] = {"api_key": api_key}
-            if timeout_s is not None:
-                http_options = getattr(types, "HttpOptions", None)
-                if http_options is not None:
-                    try:
-                        client_kwargs["http_options"] = http_options(timeout=timeout_s)
-                    except Exception:
-                        logger.debug("Gemini HttpOptions timeout not applied.", exc_info=True)
-            return genai.Client(**client_kwargs)
-
-        def _run_request() -> bytes:
-            client = _build_client()
-            try:
-                try:
-                    response = client.models.generate_content(
-                        model=model,
-                        contents=contents,
-                        config=config,
-                    )
-                except Exception as exc:
-                    logger.exception(
-                        "Gemini generate_image failed: %s request_id=%s",
-                        exc,
-                        request_id,
-                    )
-                    raise
-            finally:
-                closer = getattr(client, "close", None)
-                if callable(closer):
-                    try:
-                        closer()
-                    except Exception:
-                        pass
-
-            if not response.parts:
-                raise ValueError("No content returned from Gemini.")
-
-            for part in response.parts:
-                if part.inline_data:
-                    return part.inline_data.data
-
-            raise ValueError("No image data found in response.")
-
-        return run_with_retries(
-            func=_run_request,
-            max_retries=retry_count,
-            retry_backoff_s=retry_backoff_s,
-            request_id=request_id,
-        )
-
-    async def async_generate_image(
-        self,
-        *,
-        api_key: str,
-        prompt: str,
-        model: str,
-        image_size: Optional[str] = None,
-        aspect_ratio: Optional[str] = None,
-        image: Optional[ImageInput] = None,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> bytes:
-        return await run_sync_in_thread(
-            lambda: self.generate_image(
-                api_key=api_key,
-                prompt=prompt,
-                model=model,
-                image_size=image_size,
-                aspect_ratio=aspect_ratio,
-                image=image,
-                request_id=request_id,
-                timeout_s=timeout_s,
-                max_retries=max_retries,
-                retry_backoff_s=retry_backoff_s,
-            )
-        )
-
-    def list_models(
-        self,
-        *,
-        api_key: str,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> list[dict[str, Optional[str]]]:
-        """Return the models available to the authenticated Gemini account."""
-        if not api_key:
-            raise ValueError("api_key must be provided.")
-
-        retry_count = clamp_retries(max_retries)
-
-        def _build_client() -> genai.Client:
-            client_kwargs: dict[str, object] = {"api_key": api_key}
-            if timeout_s is not None:
-                http_options = getattr(types, "HttpOptions", None)
-                if http_options is not None:
-                    try:
-                        client_kwargs["http_options"] = http_options(timeout=timeout_s)
-                    except Exception:
-                        logger.debug("Gemini HttpOptions timeout not applied.", exc_info=True)
-            return genai.Client(**client_kwargs)
-
-        def _run_request() -> list[dict[str, Optional[str]]]:
-            models: list[dict[str, Optional[str]]] = []
-            client = _build_client()
-            try:
-                try:
-                    iterator = client.models.list()
-                except Exception as exc:
-                    logger.exception(
-                        "Gemini list models failed: %s request_id=%s",
-                        exc,
-                        request_id,
-                    )
-                    raise
-                for model in iterator:
-                    model_id = getattr(model, "name", None)
-                    if model_id is None and isinstance(model, dict):
-                        model_id = model.get("name")
-                    if not model_id:
-                        continue
-
-                    # Normalize IDs like "models/<id>" -> "<id>"
-                    if isinstance(model_id, str) and model_id.startswith("models/"):
-                        model_id = model_id.split("/", 1)[1]
-
-                    display_name = getattr(model, "display_name", None)
-                    if display_name is None and isinstance(model, dict):
-                        display_name = model.get("display_name")
-
-                    models.append({"id": model_id, "display_name": display_name})
-            finally:
-                closer = getattr(client, "close", None)
-                if callable(closer):
-                    try:
-                        closer()
-                    except Exception:
-                        pass
-
-            logger.info(
-                "Gemini list_models succeeded: count=%d request_id=%s",
-                len(models),
-                request_id,
-            )
-            return models
-
-        return run_with_retries(
-            func=_run_request,
-            max_retries=retry_count,
-            retry_backoff_s=retry_backoff_s,
-            request_id=request_id,
-        )
-
-    async def async_list_models(
-        self,
-        *,
-        api_key: str,
-        request_id: Optional[str] = None,
-        timeout_s: Optional[float] = None,
-        max_retries: Optional[int] = None,
-        retry_backoff_s: float = 0.5,
-    ) -> list[dict[str, Optional[str]]]:
-        return await run_sync_in_thread(
-            lambda: self.list_models(
-                api_key=api_key,
-                request_id=request_id,
-                timeout_s=timeout_s,
-                max_retries=max_retries,
-                retry_backoff_s=retry_backoff_s,
-            )
-        )
+        retry_count = clamp_retries(max_retries)
+
+        def _build_client() -> genai.Client:
+            client_kwargs: dict[str, object] = {"api_key": api_key}
+            if timeout_s is not None:
+                # Gemini requires at least 10s timeout if set
+                effective_timeout = max(10.0, timeout_s)
+                if effective_timeout != timeout_s:
+                    logger.warning("Gemini timeout %ss is too short, clamping to %ss.", timeout_s, effective_timeout)
+                client_kwargs["http_options"] = types.HttpOptions(timeout=effective_timeout)
+            return genai.Client(**client_kwargs)
+
+        def _run_request() -> bytes:
+            client = _build_client()
+            try:
+                try:
+                    response = client.models.generate_content(
+                        model=model,
+                        contents=contents,
+                        config=config,
+                    )
+                except Exception as exc:
+                    logger.exception(
+                        "Gemini generate_image failed: %s request_id=%s",
+                        exc,
+                        request_id,
+                    )
+                    raise
+            finally:
+                closer = getattr(client, "close", None)
+                if callable(closer):
+                    try:
+                        closer()
+                    except Exception:
+                        pass
+
+            if not response.parts:
+                raise ValueError("No content returned from Gemini.")
+
+            for part in response.parts:
+                if part.inline_data:
+                    return part.inline_data.data
+
+            raise ValueError("No image data found in response.")
+
+        return run_with_retries(
+            func=_run_request,
+            max_retries=retry_count,
+            retry_backoff_s=retry_backoff_s,
+            request_id=request_id,
+        )
+
+    async def async_generate_image(
+        self,
+        *,
+        api_key: str,
+        prompt: str,
+        model: str,
+        image_size: Optional[str] = None,
+        aspect_ratio: Optional[str] = None,
+        image: Optional[ImageInput] = None,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> bytes:
+        return await run_sync_in_thread(
+            lambda: self.generate_image(
+                api_key=api_key,
+                prompt=prompt,
+                model=model,
+                image_size=image_size,
+                aspect_ratio=aspect_ratio,
+                image=image,
+                request_id=request_id,
+                timeout_s=timeout_s,
+                max_retries=max_retries,
+                retry_backoff_s=retry_backoff_s,
+            )
+        )
+
+    def list_models(
+        self,
+        *,
+        api_key: str,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> list[dict[str, Optional[str]]]:
+        """Return the models available to the authenticated Gemini account."""
+        if not api_key:
+            raise ValueError("api_key must be provided.")
+
+        retry_count = clamp_retries(max_retries)
+
+        def _build_client() -> genai.Client:
+            client_kwargs: dict[str, object] = {"api_key": api_key}
+            if timeout_s is not None:
+                # Gemini requires at least 10s timeout if set
+                effective_timeout = max(10.0, timeout_s)
+                if effective_timeout != timeout_s:
+                    logger.warning("Gemini timeout %ss is too short, clamping to %ss.", timeout_s, effective_timeout)
+                client_kwargs["http_options"] = types.HttpOptions(timeout=effective_timeout)
+            return genai.Client(**client_kwargs)
+
+        def _run_request() -> list[dict[str, Optional[str]]]:
+            models: list[dict[str, Optional[str]]] = []
+            client = _build_client()
+            try:
+                try:
+                    iterator = client.models.list()
+                except Exception as exc:
+                    logger.exception(
+                        "Gemini list models failed: %s request_id=%s",
+                        exc,
+                        request_id,
+                    )
+                    raise
+                for model in iterator:
+                    model_id = getattr(model, "name", None)
+                    if model_id is None and isinstance(model, dict):
+                        model_id = model.get("name")
+                    if not model_id:
+                        continue
+
+                    # Normalize IDs like "models/<id>" -> "<id>"
+                    if isinstance(model_id, str) and model_id.startswith("models/"):
+                        model_id = model_id.split("/", 1)[1]
+
+                    display_name = getattr(model, "display_name", None)
+                    if display_name is None and isinstance(model, dict):
+                        display_name = model.get("display_name")
+
+                    models.append({"id": model_id, "display_name": display_name})
+            finally:
+                closer = getattr(client, "close", None)
+                if callable(closer):
+                    try:
+                        closer()
+                    except Exception:
+                        pass
+
+            logger.info(
+                "Gemini list_models succeeded: count=%d request_id=%s",
+                len(models),
+                request_id,
+            )
+            return models
+
+        return run_with_retries(
+            func=_run_request,
+            max_retries=retry_count,
+            retry_backoff_s=retry_backoff_s,
+            request_id=request_id,
+        )
+
+    async def async_list_models(
+        self,
+        *,
+        api_key: str,
+        request_id: Optional[str] = None,
+        timeout_s: Optional[float] = None,
+        max_retries: Optional[int] = None,
+        retry_backoff_s: float = 0.5,
+    ) -> list[dict[str, Optional[str]]]:
+        return await run_sync_in_thread(
+            lambda: self.list_models(
+                api_key=api_key,
+                request_id=request_id,
+                timeout_s=timeout_s,
+                max_retries=max_retries,
+                retry_backoff_s=retry_backoff_s,
+            )
+        )
 
     @staticmethod
     def _to_image_part(image: ImageInput) -> types.Part: