google-genai 1.7.0__py3-none-any.whl → 1.53.0__py3-none-any.whl

google/genai/chats.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Google LLC
+# Copyright 2025 Google LLC
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,13 +13,14 @@
 # limitations under the License.
 #
 
+from collections.abc import Iterator
 import sys
 from typing import AsyncIterator, Awaitable, Optional, Union, get_args
 
 from . import _transformers as t
 from . import types
 from .models import AsyncModels, Models
-from .types import Content, GenerateContentConfigOrDict, GenerateContentResponse, Part, PartUnionDict
+from .types import Content, ContentOrDict, GenerateContentConfigOrDict, GenerateContentResponse, Part, PartUnionDict
 
 
 if sys.version_info >= (3, 10):
@@ -62,13 +63,8 @@ def _extract_curated_history(
   """Extracts the curated (valid) history from a comprehensive history.
 
   The comprehensive history contains all turns (user input and model responses),
-  including any invalid or rejected model outputs.  This function filters
-  that history to return only the valid turns.
-
-  A "turn" starts with one user input (a single content) and then follows by
-  corresponding model response (which may consist of multiple contents).
-  Turns are assumed to alternate: user input, model output, user input, model
-  output, etc.
+  including any invalid or rejected model outputs. This function filters that
+  history to return only the valid turns.
 
   Args:
       comprehensive_history: A list representing the complete chat history.
@@ -83,8 +79,6 @@ def _extract_curated_history(
   length = len(comprehensive_history)
   i = 0
   current_input = comprehensive_history[i]
-  if current_input.role != "user":
-    raise ValueError("History must start with a user turn.")
   while i < length:
     if comprehensive_history[i].role not in ["user", "model"]:
       raise ValueError(
@@ -93,6 +87,7 @@ def _extract_curated_history(
 
     if comprehensive_history[i].role == "user":
       current_input = comprehensive_history[i]
+      curated_history.append(current_input)
       i += 1
     else:
       current_output = []
@@ -103,8 +98,9 @@ def _extract_curated_history(
           is_valid = False
         i += 1
       if is_valid:
-        curated_history.append(current_input)
         curated_history.extend(current_output)
+      elif curated_history:
+        curated_history.pop()
   return curated_history
 
 
@@ -116,14 +112,21 @@ class _BaseChat:
       *,
       model: str,
       config: Optional[GenerateContentConfigOrDict] = None,
-      history: list[Content],
+      history: list[ContentOrDict],
   ):
     self._model = model
     self._config = config
-    self._comprehensive_history = history
+    content_models = []
+    for content in history:
+      if not isinstance(content, Content):
+        content_model = Content.model_validate(content)
+      else:
+        content_model = content
+      content_models.append(content_model)
+    self._comprehensive_history = content_models
     """Comprehensive history is the full history of the chat, including turns of the invalid contents from the model and their associated inputs.
     """
-    self._curated_history = _extract_curated_history(history)
+    self._curated_history = _extract_curated_history(content_models)
     """Curated history is the set of valid turns that will be used in the subsequent send requests.
     """
 
@@ -133,7 +136,7 @@ class _BaseChat:
       model_output: list[Content],
       automatic_function_calling_history: list[Content],
       is_valid: bool,
-  ):
+  ) -> None:
     """Records the chat history.
 
     Maintaining both comprehensive and curated histories.
@@ -152,7 +155,7 @@ class _BaseChat:
         # Because the AFC input contains the entire curated chat history in
         # addition to the new user input, we need to truncate the AFC history
         # to deduplicate the existing chat history.
-        automatic_function_calling_history[len(self._curated_history):]
+        automatic_function_calling_history[len(self._curated_history) :]
         if automatic_function_calling_history
         else [user_input]
     )
@@ -210,7 +213,7 @@ class Chat(_BaseChat):
       modules: Models,
       model: str,
       config: Optional[GenerateContentConfigOrDict] = None,
-      history: list[Content],
+      history: list[ContentOrDict],
   ):
     self._modules = modules
     super().__init__(
@@ -238,7 +241,7 @@ class Chat(_BaseChat):
 
     .. code-block:: python
 
-      chat = client.chats.create(model='gemini-1.5-flash')
+      chat = client.chats.create(model='gemini-2.0-flash')
       response = chat.send_message('tell me a story')
     """
 
@@ -247,7 +250,7 @@ class Chat(_BaseChat):
           f"Message must be a valid part type: {types.PartUnion} or"
           f" {types.PartUnionDict}, got {type(message)}"
       )
-    input_content = t.t_content(self._modules._api_client, message)
+    input_content = t.t_content(message)
     response = self._modules.generate_content(
         model=self._model,
         contents=self._curated_history + [input_content],  # type: ignore[arg-type]
@@ -275,7 +278,7 @@ class Chat(_BaseChat):
       self,
       message: Union[list[PartUnionDict], PartUnionDict],
       config: Optional[GenerateContentConfigOrDict] = None,
-  ):
+  ) -> Iterator[GenerateContentResponse]:
     """Sends the conversation history with the additional message and yields the model's response in chunks.
 
     Args:
@@ -290,7 +293,7 @@ class Chat(_BaseChat):
 
     .. code-block:: python
 
-      chat = client.chats.create(model='gemini-1.5-flash')
+      chat = client.chats.create(model='gemini-2.0-flash')
       for chunk in chat.send_message_stream('tell me a story'):
         print(chunk.text)
     """
@@ -300,7 +303,7 @@ class Chat(_BaseChat):
           f"Message must be a valid part type: {types.PartUnion} or"
           f" {types.PartUnionDict}, got {type(message)}"
       )
-    input_content = t.t_content(self._modules._api_client, message)
+    input_content = t.t_content(message)
     output_contents = []
     finish_reason = None
     is_valid = True
@@ -320,7 +323,7 @@ class Chat(_BaseChat):
         yield chunk
       automatic_function_calling_history = (
           chunk.automatic_function_calling_history
-          if chunk.automatic_function_calling_history
+          if chunk is not None and chunk.automatic_function_calling_history
           else []
       )
       self.record_history(
@@ -344,7 +347,7 @@ class Chats:
       *,
       model: str,
       config: Optional[GenerateContentConfigOrDict] = None,
-      history: Optional[list[Content]] = None,
+      history: Optional[list[ContentOrDict]] = None,
   ) -> Chat:
     """Creates a new chat session.
 
@@ -373,7 +376,7 @@ class AsyncChat(_BaseChat):
       modules: AsyncModels,
       model: str,
       config: Optional[GenerateContentConfigOrDict] = None,
-      history: list[Content],
+      history: list[ContentOrDict],
   ):
     self._modules = modules
     super().__init__(
@@ -401,7 +404,7 @@ class AsyncChat(_BaseChat):
 
     .. code-block:: python
 
-      chat = client.aio.chats.create(model='gemini-1.5-flash')
+      chat = client.aio.chats.create(model='gemini-2.0-flash')
       response = await chat.send_message('tell me a story')
     """
     if not _is_part_type(message):
@@ -409,7 +412,7 @@ class AsyncChat(_BaseChat):
           f"Message must be a valid part type: {types.PartUnion} or"
           f" {types.PartUnionDict}, got {type(message)}"
       )
-    input_content = t.t_content(self._modules._api_client, message)
+    input_content = t.t_content(message)
     response = await self._modules.generate_content(
         model=self._model,
         contents=self._curated_history + [input_content],  # type: ignore[arg-type]
@@ -437,7 +440,7 @@ class AsyncChat(_BaseChat):
       self,
       message: Union[list[PartUnionDict], PartUnionDict],
       config: Optional[GenerateContentConfigOrDict] = None,
-  ) -> Awaitable[AsyncIterator[GenerateContentResponse]]:
+  ) -> AsyncIterator[GenerateContentResponse]:
     """Sends the conversation history with the additional message and yields the model's response in chunks.
 
     Args:
@@ -451,7 +454,8 @@ class AsyncChat(_BaseChat):
     Usage:
 
     .. code-block:: python
-      chat = client.aio.chats.create(model='gemini-1.5-flash')
+
+      chat = client.aio.chats.create(model='gemini-2.0-flash')
       async for chunk in await chat.send_message_stream('tell me a story'):
         print(chunk.text)
     """
@@ -461,16 +465,16 @@ class AsyncChat(_BaseChat):
           f"Message must be a valid part type: {types.PartUnion} or"
           f" {types.PartUnionDict}, got {type(message)}"
       )
-    input_content = t.t_content(self._modules._api_client, message)
+    input_content = t.t_content(message)
 
-    async def async_generator():
+    async def async_generator():  # type: ignore[no-untyped-def]
       output_contents = []
       finish_reason = None
       is_valid = True
       chunk = None
-      async for chunk in await self._modules.generate_content_stream(
+      async for chunk in await self._modules.generate_content_stream(  # type: ignore[attr-defined]
           model=self._model,
-          contents=self._curated_history + [input_content],
+          contents=self._curated_history + [input_content],  # type: ignore[arg-type]
           config=config if config else self._config,
       ):
         if not _validate_response(chunk):
@@ -481,13 +485,19 @@ class AsyncChat(_BaseChat):
           finish_reason = chunk.candidates[0].finish_reason
         yield chunk
 
+      if not output_contents or finish_reason is None:
+        is_valid = False
+
       self.record_history(
           user_input=input_content,
           model_output=output_contents,
-          automatic_function_calling_history=chunk.automatic_function_calling_history,
-          is_valid=is_valid and output_contents and finish_reason,
+          automatic_function_calling_history=chunk.automatic_function_calling_history
+          if chunk is not None and chunk.automatic_function_calling_history
+          else [],
+          is_valid=is_valid,
       )
-    return async_generator()
+
+    return async_generator()  # type: ignore[no-untyped-call, no-any-return]
 
 
 class AsyncChats:
@@ -501,7 +511,7 @@ class AsyncChats:
       *,
       model: str,
       config: Optional[GenerateContentConfigOrDict] = None,
-      history: Optional[list[Content]] = None,
+      history: Optional[list[ContentOrDict]] = None,
   ) -> AsyncChat:
     """Creates a new chat session.
 

google/genai/client.py

@@ -1,4 +1,4 @@
-# Copyright 2024 Google LLC
+# Copyright 2025 Google LLC
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -13,22 +13,28 @@
 # limitations under the License.
 #
 
+import asyncio
 import os
+from types import TracebackType
 from typing import Optional, Union
 
 import google.auth
 import pydantic
 
-from ._api_client import BaseApiClient, HttpOptions, HttpOptionsDict
+from ._api_client import BaseApiClient
+from ._base_url import get_base_url
 from ._replay_api_client import ReplayApiClient
 from .batches import AsyncBatches, Batches
 from .caches import AsyncCaches, Caches
 from .chats import AsyncChats, Chats
+from .file_search_stores import AsyncFileSearchStores, FileSearchStores
 from .files import AsyncFiles, Files
 from .live import AsyncLive
 from .models import AsyncModels, Models
 from .operations import AsyncOperations, Operations
+from .tokens import AsyncTokens, Tokens
 from .tunings import AsyncTunings, Tunings
+from .types import HttpOptions, HttpOptionsDict, HttpRetryOptions
 
 
 class AsyncClient:
@@ -42,7 +48,9 @@ class AsyncClient:
     self._caches = AsyncCaches(self._api_client)
     self._batches = AsyncBatches(self._api_client)
     self._files = AsyncFiles(self._api_client)
+    self._file_search_stores = AsyncFileSearchStores(self._api_client)
     self._live = AsyncLive(self._api_client)
+    self._tokens = AsyncTokens(self._api_client)
     self._operations = AsyncOperations(self._api_client)
 
   @property
@@ -57,6 +65,10 @@ class AsyncClient:
   def caches(self) -> AsyncCaches:
     return self._caches
 
+  @property
+  def file_search_stores(self) -> AsyncFileSearchStores:
+    return self._file_search_stores
+
   @property
   def batches(self) -> AsyncBatches:
     return self._batches
@@ -73,10 +85,59 @@ class AsyncClient:
   def live(self) -> AsyncLive:
     return self._live
 
+  @property
+  def auth_tokens(self) -> AsyncTokens:
+    return self._tokens
+
   @property
   def operations(self) -> AsyncOperations:
     return self._operations
 
+  async def aclose(self) -> None:
+    """Closes the async client explicitly.
+
+    However, it doesn't close the sync client, which can be closed using the
+    Client.close() method or using the context manager.
+
+    Usage:
+    .. code-block:: python
+
+      from google.genai import Client
+
+      async_client = Client(
+          vertexai=True, project='my-project-id', location='us-central1'
+      ).aio
+      response_1 = await async_client.models.generate_content(
+          model='gemini-2.0-flash',
+          contents='Hello World',
+      )
+      response_2 = await async_client.models.generate_content(
+          model='gemini-2.0-flash',
+          contents='Hello World',
+      )
+      # Close the client to release resources.
+      await async_client.aclose()
+    """
+    await self._api_client.aclose()
+
+  async def __aenter__(self) -> 'AsyncClient':
+    return self
+
+  async def __aexit__(
+      self,
+      exc_type: Optional[Exception],
+      exc_value: Optional[Exception],
+      traceback: Optional[TracebackType],
+  ) -> None:
+    await self.aclose()
+
+  def __del__(self) -> None:
+    try:
+      asyncio.get_running_loop().create_task(self.aclose())
+    except Exception:
+      pass
+
+
 class DebugConfig(pydantic.BaseModel):
   """Configuration options that change client network behavior when testing."""
 
@@ -113,26 +174,29 @@ class Client:
   Attributes:
     api_key: The `API key <https://ai.google.dev/gemini-api/docs/api-key>`_ to
       use for authentication. Applies to the Gemini Developer API only.
-    vertexai: Indicates whether the client should use the Vertex AI
-      API endpoints. Defaults to False (uses Gemini Developer API endpoints).
+    vertexai: Indicates whether the client should use the Vertex AI API
+      endpoints. Defaults to False (uses Gemini Developer API endpoints).
       Applies to the Vertex AI API only.
     credentials: The credentials to use for authentication when calling the
       Vertex AI APIs. Credentials can be obtained from environment variables and
-      default credentials. For more information, see
-      `Set up Application Default Credentials
+      default credentials. For more information, see `Set up Application Default
+      Credentials
       <https://cloud.google.com/docs/authentication/provide-credentials-adc>`_.
       Applies to the Vertex AI API only.
-    project: The `Google Cloud project ID <https://cloud.google.com/vertex-ai/docs/start/cloud-environment>`_ to
-      use for quota. Can be obtained from environment variables (for example,
+    project: The `Google Cloud project ID
+      <https://cloud.google.com/vertex-ai/docs/start/cloud-environment>`_ to use
+      for quota. Can be obtained from environment variables (for example,
       ``GOOGLE_CLOUD_PROJECT``). Applies to the Vertex AI API only.
-    location: The `location <https://cloud.google.com/vertex-ai/generative-ai/docs/learn/locations>`_
+      Find your `Google Cloud project ID <https://cloud.google.com/resource-manager/docs/creating-managing-projects#identifying_projects>`_.
+    location: The `location
+      <https://cloud.google.com/vertex-ai/generative-ai/docs/learn/locations>`_
       to send API requests to (for example, ``us-central1``). Can be obtained
       from environment variables. Applies to the Vertex AI API only.
     debug_config: Config settings that control network behavior of the client.
       This is typically used when running test code.
     http_options: Http options to use for the client. These options will be
-      applied to all requests made by the client. Example usage:
-      `client = genai.Client(http_options=types.HttpOptions(api_version='v1'))`.
+      applied to all requests made by the client. Example usage: `client =
+      genai.Client(http_options=types.HttpOptions(api_version='v1'))`.
 
   Usage for the Gemini Developer API:
 
@@ -194,6 +258,15 @@ class Client:
     """
 
     self._debug_config = debug_config or DebugConfig()
+    if isinstance(http_options, dict):
+      http_options = HttpOptions(**http_options)
+
+    base_url = get_base_url(vertexai or False, http_options)
+    if base_url:
+      if http_options:
+        http_options.base_url = base_url
+      else:
+        http_options = HttpOptions(base_url=base_url)
 
     self._api_client = self._get_api_client(
         vertexai=vertexai,
@@ -209,8 +282,10 @@ class Client:
     self._models = Models(self._api_client)
     self._tunings = Tunings(self._api_client)
     self._caches = Caches(self._api_client)
+    self._file_search_stores = FileSearchStores(self._api_client)
     self._batches = Batches(self._api_client)
     self._files = Files(self._api_client)
+    self._tokens = Tokens(self._api_client)
     self._operations = Operations(self._api_client)
 
   @staticmethod
@@ -222,17 +297,17 @@ class Client:
       location: Optional[str] = None,
       debug_config: Optional[DebugConfig] = None,
       http_options: Optional[HttpOptions] = None,
-  ):
+  ) -> BaseApiClient:
     if debug_config and debug_config.client_mode in [
         'record',
         'replay',
         'auto',
     ]:
       return ReplayApiClient(
-          mode=debug_config.client_mode,
-          replay_id=debug_config.replay_id,
+          mode=debug_config.client_mode,  # type: ignore[arg-type]
+          replay_id=debug_config.replay_id,  # type: ignore[arg-type]
           replays_directory=debug_config.replays_directory,
-          vertexai=vertexai,
+          vertexai=vertexai,  # type: ignore[arg-type]
           api_key=api_key,
           credentials=credentials,
           project=project,
@@ -269,6 +344,10 @@ class Client:
   def caches(self) -> Caches:
     return self._caches
 
+  @property
+  def file_search_stores(self) -> FileSearchStores:
+    return self._file_search_stores
+
   @property
   def batches(self) -> Batches:
     return self._batches
@@ -277,6 +356,10 @@ class Client:
   def files(self) -> Files:
     return self._files
 
+  @property
+  def auth_tokens(self) -> Tokens:
+    return self._tokens
+
   @property
   def operations(self) -> Operations:
     return self._operations
@@ -285,3 +368,47 @@ class Client:
   def vertexai(self) -> bool:
     """Returns whether the client is using the Vertex AI API."""
     return self._api_client.vertexai or False
+
+  def close(self) -> None:
+    """Closes the synchronous client explicitly.
+
+    However, it doesn't close the async client, which can be closed using the
+    Client.aio.aclose() method or using the async context manager.
+
+    Usage:
+    .. code-block:: python
+
+      from google.genai import Client
+
+      client = Client(
+          vertexai=True, project='my-project-id', location='us-central1'
+      )
+      response_1 = client.models.generate_content(
+          model='gemini-2.0-flash',
+          contents='Hello World',
+      )
+      response_2 = client.models.generate_content(
+          model='gemini-2.0-flash',
+          contents='Hello World',
+      )
+      # Close the client to release resources.
+      client.close()
+    """
+    self._api_client.close()
+
+  def __enter__(self) -> 'Client':
+    return self
+
+  def __exit__(
+      self,
+      exc_type: Optional[Exception],
+      exc_value: Optional[Exception],
+      traceback: Optional[TracebackType],
+  ) -> None:
+    self.close()
+
+  def __del__(self) -> None:
+    try:
+      self.close()
+    except Exception:
+      pass