pygpt-net 2.6.30__py3-none-any.whl → 2.6.32__py3-none-any.whl

pygpt_net/provider/api/google/__init__.py

@@ -6,13 +6,15 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.28 20:00:00                  #
+# Updated Date: 2025.09.01 23:00:00                  #
 # ================================================== #
 
+import os
 from typing import Optional, Dict, Any
 
 from google.genai import types as gtypes
 from google import genai
+
 from pygpt_net.core.types import (
     MODE_ASSISTANT,
     MODE_AUDIO,
@@ -29,7 +31,8 @@ from .vision import Vision
 from .tools import Tools
 from .audio import Audio
 from .image import Image
-
+from .realtime import Realtime
+from .video import Video
 
 class ApiGoogle:
     def __init__(self, window=None):
@@ -44,6 +47,8 @@ class ApiGoogle:
         self.tools = Tools(window)
         self.audio = Audio(window)
         self.image = Image(window)
+        self.realtime = Realtime(window)
+        self.video = Video(window)
         self.client: Optional[genai.Client] = None
         self.locked = False
         self.last_client_args: Optional[Dict[str, Any]] = None
@@ -64,20 +69,56 @@ class ApiGoogle:
             model = ModelItem()
             model.provider = "google"
         args = self.window.core.models.prepare_client_args(mode, model)
+        config = self.window.core.config
+
         filtered = {}
         if args.get("api_key"):
             filtered["api_key"] = args["api_key"]
+
+        # setup VertexAI
+        use_vertex = False
+        if config.get("api_native_google.use_vertex", False):
+            use_vertex = True
+            os.environ["GOOGLE_GENAI_USE_VERTEXAI"] = "1"
+            os.environ["GOOGLE_CLOUD_PROJECT"] = config.get("api_native_google.cloud_project", "")
+            os.environ["GOOGLE_CLOUD_LOCATION"] = config.get("api_native_google.cloud_location", "us-central1")
+            if config.get("api_native_google.app_credentials", ""):
+                os.environ["GOOGLE_APPLICATION_CREDENTIALS"] = config.get("api_native_google.app_credentials", "")
+        else:
+            if os.environ.get("GOOGLE_GENAI_USE_VERTEXAI"):
+                del os.environ["GOOGLE_GENAI_USE_VERTEXAI"]
+            if os.environ.get("GOOGLE_CLOUD_PROJECT"):
+                del os.environ["GOOGLE_CLOUD_PROJECT"]
+            if os.environ.get("GOOGLE_CLOUD_LOCATION"):
+                del os.environ["GOOGLE_CLOUD_LOCATION"]
+            if os.environ.get("GOOGLE_APPLICATION_CREDENTIALS"):
+                del os.environ["GOOGLE_APPLICATION_CREDENTIALS"]
+
+        # append VertexAI params to client args
+        if use_vertex:
+            filtered["vertexai"] = True
+            filtered["project"] = os.environ.get("GOOGLE_CLOUD_PROJECT")
+            filtered["location"] = os.environ.get("GOOGLE_CLOUD_LOCATION", "us-central1")
+            # filtered["http_options"] = gtypes.HttpOptions(api_version="v1")
+
         if self.client is None or self.last_client_args != filtered:
             self.client = genai.Client(**filtered)
         self.last_client_args = filtered
+
         return self.client
 
-    def call(self, context: BridgeContext, extra: dict = None) -> bool:
+    def call(
+            self,
+            context: BridgeContext,
+            extra: dict = None,
+            rt_signals = None
+    ) -> bool:
         """
         Make an API call to Google GenAI
 
         :param context: BridgeContext
         :param extra: Extra parameters
+        :param rt_signals: Realtime signals for audio streaming
         :return: True if successful, False otherwise
         """
         mode = context.mode
@@ -94,13 +135,28 @@ class ApiGoogle:
         response = None
 
         if mode in [MODE_COMPLETION, MODE_CHAT, MODE_AUDIO, MODE_RESEARCH]:
+
+            # Live API for audio streaming
+            if mode == MODE_AUDIO and stream:
+                is_realtime = self.realtime.begin(
+                    context=context,
+                    model=model,
+                    extra=extra or {},
+                    rt_signals=rt_signals
+                )
+                if is_realtime:
+                    return True
+
             response = self.chat.send(context=context, extra=extra)
             used_tokens = self.chat.get_used_tokens()
             if ctx:
                 self.vision.append_images(ctx)
 
         elif mode == MODE_IMAGE:
-            return self.image.generate(context=context, extra=extra)
+            if context.model.is_video_output():
+                return self.video.generate(context=context, extra=extra)  # veo, etc.
+            else:
+                return self.image.generate(context=context, extra=extra) # imagen, etc.
 
         elif mode == MODE_ASSISTANT:
             return False  # not implemented for Google
@@ -135,7 +191,11 @@ class ApiGoogle:
                 pass
         return True
 
-    def quick_call(self, context: BridgeContext, extra: dict = None) -> str:
+    def quick_call(
+            self,
+            context: BridgeContext,
+            extra: dict = None
+    ) -> str:
         """
         Make a quick API call to Google GenAI and return the output text
 
@@ -206,9 +266,9 @@ class ApiGoogle:
     def build_remote_tools(self, model: ModelItem = None) -> list:
         """
         Build Google GenAI remote tools based on config flags.
-        - google_tool_search: enables grounding via Google Search (Gemini 2.x)
+        - remote_tools.google.web_search: enables grounding via Google Search (Gemini 2.x)
           or GoogleSearchRetrieval (Gemini 1.5 fallback).
-        - google_tool_code_execution: enables code execution tool.
+        - remote_tools.google.code_interpreter: enables code execution tool.
 
         Returns a list of gtypes.Tool objects (can be empty).
 
@@ -242,6 +302,15 @@ class ApiGoogle:
             except Exception as e:
                 self.window.core.debug.log(e)
 
+        # URL Context tool
+        if cfg.get("remote_tools.google.url_ctx") and "image" not in model.id:
+            try:
+                # Supported on Gemini 2.x+ models (not on 1.5)
+                if not model_id.startswith("gemini-1.5") and not model_id.startswith("models/gemini-1.5"):
+                    tools.append(gtypes.Tool(url_context=gtypes.UrlContext))
+            except Exception as e:
+                self.window.core.debug.log(e)
+
         return tools
 
 

pygpt_net/provider/api/google/audio.py

@@ -24,6 +24,8 @@ class Audio:
         Audio helpers for Google GenAI.
         - Build audio input parts for requests
         - Convert Google PCM output to WAV (base64) for UI compatibility
+
+        :param window: Window instance
         """
         self.window = window
 
@@ -103,7 +105,12 @@ class Audio:
 
     @staticmethod
     def _ensure_bytes(data) -> Optional[bytes]:
-        """Return raw bytes from inline_data.data (bytes or base64 string)."""
+        """
+        Return raw bytes from inline_data.data (bytes or base64 string).
+
+        :param data: bytes or base64 string
+        :return: bytes or None
+        """
         try:
             if isinstance(data, (bytes, bytearray)):
                 return bytes(data)

pygpt_net/provider/api/google/chat.py

@@ -29,9 +29,17 @@ class Chat:
         self.window = window
         self.input_tokens = 0
 
-    def send(self, context: BridgeContext, extra: Optional[Dict[str, Any]] = None):
+    def send(
+            self,
+            context: BridgeContext,
+            extra: Optional[Dict[str, Any]] = None
+    ):
         """
         Call Google GenAI for chat / multimodal / audio.
+
+        :param context: BridgeContext with prompt, model, history, mode, etc.
+        :param extra: Extra parameters (not used currently)
+        :return: Response object or generator (if streaming)
         """
         prompt = context.prompt
         stream = context.stream
@@ -110,9 +118,13 @@ class Chat:
         # Tools -> merge app-defined tools with remote tools
         base_tools = self.window.core.api.google.tools.prepare(model, functions)
         remote_tools = self.window.core.api.google.build_remote_tools(model)
+
+        # Check tools compatibility
         if base_tools:
-            remote_tools = [] # do not mix local and remote tools
+            remote_tools = [] # remote tools are not allowed if function calling is used
         tools = (base_tools or []) + (remote_tools or [])
+        if "-image" in model.id:
+            tools = None  # function calling is not supported for image models
 
         # Sampling
         temperature = self.window.core.config.get('temperature')
@@ -144,7 +156,7 @@ class Chat:
             # Voice selection (case-sensitive name)
             voice_name = "Kore"
             try:
-                tmp = self.window.core.plugins.get_option("audio_output", "google_voice_native")
+                tmp = self.window.core.plugins.get_option("audio_output", "google_genai_tts_voice")
                 if tmp:
                     name = str(tmp).strip()
                     mapping = {"kore": "Kore", "puck": "Puck", "charon": "Charon", "verse": "Verse", "legend": "Legend"}
@@ -169,9 +181,17 @@ class Chat:
         else:
             return client.models.generate_content(**params)
 
-    def unpack_response(self, mode: str, response, ctx: CtxItem):
+    def unpack_response(
+            self,
+            mode: str,
+            response, ctx: CtxItem
+    ):
         """
         Unpack non-streaming response from Google GenAI and set context.
+
+        :param mode: MODE_CHAT or MODE_AUDIO
+        :param response: Response object
+        :param ctx: CtxItem to set output, audio_output, tokens, tool_calls
         """
         if mode == MODE_AUDIO:
             # Prefer audio if present
@@ -229,6 +249,11 @@ class Chat:
     def extract_text(self, response) -> str:
         """
         Extract output text.
+
+        Prefer response.text (Python SDK), then fallback to parts[].text.
+
+        :param response: Response object
+        :return: Extracted text
         """
         txt = getattr(response, "text", None) or getattr(response, "output_text", None)
         if txt:
@@ -332,11 +357,17 @@ class Chat:
 
         return out
 
-    def _extract_inline_images_and_links(self, response, ctx: CtxItem) -> None:
+    def _extract_inline_images_and_links(
+            self,
+            response, ctx: CtxItem
+    ) -> None:
         """
         Extract inline image parts (Gemini image output) and file links.
         - Saves inline_data (image/*) bytes to files and appends paths to ctx.images.
         - Appends HTTP(S) image URIs from file_data to ctx.urls.
+
+        :param response: Response object
+        :param ctx: CtxItem to set images and urls
         """
         images: list[str] = []
         urls: list[str] = []
@@ -386,7 +417,12 @@ class Chat:
 
     @staticmethod
     def _ensure_bytes(data) -> bytes | None:
-        """Return raw bytes from SDK part.inline_data.data which can be bytes or base64 string."""
+        """
+        Return raw bytes from SDK part.inline_data.data which can be bytes or base64 string.
+
+        :param data: bytes or str
+        :return: bytes or None
+        """
         try:
             if isinstance(data, (bytes, bytearray)):
                 return bytes(data)
@@ -545,6 +581,9 @@ class Chat:
         Heuristic check if the model supports native TTS.
         - Official TTS models contain '-tts' in id (e.g. 'gemini-2.5-flash-preview-tts').
         - Future/preview names may contain 'native-audio'.
+
+        :param model_id: Model ID
+        :return: True if supports TTS, False otherwise
         """
         if not model_id:
             return False