pygpt-net 2.6.30__py3-none-any.whl → 2.6.31__py3-none-any.whl

pygpt_net/data/locale/plugin.audio_output.en.ini

@@ -15,6 +15,10 @@ eleven_labs_voice.description = Specify the Voice ID.
 eleven_labs_voice.label = Voice ID
 google_api_key.description = You can obtain your own API key here: https://console.cloud.google.com/apis/library/texttospeech.googleapis.com
 google_api_key.label = Google Cloud Text-to-speech API Key
+google_genai_tts_model.description = Specify Gemini TTS model, e.g.: gemini-2.5-flash-preview-tts or gemini-2.5-pro-preview-tts
+google_genai_tts_model.label = Model
+google_genai_tts_voice.description = Specify voice, e.g.: Puck, Kore, Charon, Leda, Zephyr... (case-sensitive)
+google_genai_tts_voice.label = Voice
 google_lang.description = Specify the language code.
 google_lang.label = Language code
 google_voice.description = Specify the voice.

pygpt_net/plugin/audio_input/plugin.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2024.11.26 19:00:00                  #
+# Updated Date: 2025.08.31 23:00:00                  #
 # ================================================== #
 
 import os
@@ -23,6 +23,7 @@ from pygpt_net.utils import trans
 from .config import Config
 from .worker import Worker
 from .simple import Simple
+from ...core.types import MODE_AUDIO
 
 
 class Plugin(BasePlugin):
@@ -124,13 +125,31 @@ class Plugin(BasePlugin):
             words = [x.strip() for x in words]  # remove white-spaces
         return words
 
-    def toggle_recording_simple(self):
+    def toggle_recording_simple(
+            self,
+            state: bool = None,
+            auto: bool = False
+    ):
         """
         Event: AUDIO_INPUT_RECORD_TOGGLE
 
         Toggle recording
+
+        :param state: state to set
+        :param auto: True if called automatically (not by user)
+        """
+        if self.window.controller.realtime.is_enabled():
+            self.handler_simple.toggle_realtime(state=state, auto=auto)
+            return
+        self.handler_simple.toggle_recording(state=state)
+
+    def is_recording(self) -> bool:
         """
-        self.handler_simple.toggle_recording()
+        Check if is recording (simple mode)
+
+        :return: True if is recording
+        """
+        return self.handler_simple.is_recording
 
     def toggle_speech(self, state: bool):
         """
@@ -214,7 +233,9 @@ class Plugin(BasePlugin):
             self.toggle_speech(data['value'])
 
         elif name == Event.AUDIO_INPUT_RECORD_TOGGLE:
-            self.toggle_recording_simple()
+            state = data['state'] if 'value' in data else None
+            auto = data['auto'] if 'auto' in data else False
+            self.toggle_recording_simple(state=state, auto=auto)
 
         elif name == Event.AUDIO_INPUT_STOP:
             self.on_stop()
@@ -492,6 +513,18 @@ class Plugin(BasePlugin):
                 self.window.dispatch(event)  # send text, input clear in send method
                 self.set_status('')
 
+    def handle_realtime_stopped(self):
+        """Handle realtime stopped"""
+        context = BridgeContext()
+        context.prompt = "..."
+        extra = {}
+        event = KernelEvent(KernelEvent.INPUT_SYSTEM, {
+            'context': context,
+            'extra': extra,
+        })
+        self.window.dispatch(event)  # send text, input clear in send method
+        self.set_status('')
+
     @Slot(object)
     def handle_status(self, data: str):
         """

pygpt_net/plugin/audio_input/simple.py

@@ -6,14 +6,14 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.27 07:00:00                  #
+# Updated Date: 2025.08.31 23:00:00                  #
 # ================================================== #
 
 import os
 
 from PySide6.QtCore import QTimer
 
-from pygpt_net.core.events import AppEvent
+from pygpt_net.core.events import AppEvent, RealtimeEvent
 from pygpt_net.core.tabs.tab import Tab
 from pygpt_net.utils import trans
 
@@ -32,8 +32,46 @@ class Simple:
         self.is_recording = False
         self.timer = None
 
-    def toggle_recording(self):
-        """Toggle recording"""
+    def toggle_realtime(
+            self,
+            state: bool = None,
+            auto: bool = False
+    ):
+        """
+        Toggle recording
+
+        :param state: True to start recording, False to stop recording, None to toggle
+        :param auto: True if called automatically (not by user)
+        """
+        if state is not None:
+            if state and not self.is_recording:
+                self.start_recording(realtime=True)
+            elif not state:
+                self.force_stop()
+            else:
+                self.force_stop()
+            return
+        if self.is_recording:
+            self.stop_recording(realtime=True)
+            if not auto:
+                self.plugin.window.dispatch(RealtimeEvent(RealtimeEvent.RT_INPUT_AUDIO_MANUAL_STOP))
+        else:
+            self.start_recording(realtime=True)
+            if not auto:
+                self.plugin.window.dispatch(RealtimeEvent(RealtimeEvent.RT_INPUT_AUDIO_MANUAL_START))
+
+    def toggle_recording(self, state: bool = None):
+        """
+        Toggle recording
+
+        :param state: True to start recording, False to stop recording, None to toggle
+        """
+        if state is not None:
+            if state and not self.is_recording:
+                self.start_recording()
+            elif not state:
+                self.force_stop()
+            return
         if self.is_recording:
             self.stop_recording()
         else:
@@ -51,11 +89,12 @@ class Simple:
         """Stop timeout"""
         self.stop_recording(timeout=True)
 
-    def start_recording(self, force: bool = False):
+    def start_recording(self, force: bool = False, realtime: bool = False):
         """
         Start recording
 
         :param force: True to force recording
+        :param realtime: True if called from realtime callback
         """
         # display snap warning if not displayed yet
         if (not self.plugin.window.core.config.get("audio.input.snap", False)
@@ -89,7 +128,7 @@ class Simple:
             # disable in continuous mode
             timeout = int(self.plugin.window.core.config.get('audio.input.timeout', 120) or 0) # get timeout
             timeout_continuous = self.plugin.window.core.config.get('audio.input.timeout.continuous', False) # enable continuous timeout
-            if timeout > 0:
+            if timeout > 0 and not realtime:
                 if self.timer is None and (not continuous_enabled or timeout_continuous):
                     self.timer = QTimer()
                     self.timer.timeout.connect(self.stop_timeout)
@@ -119,11 +158,12 @@ class Simple:
                 )
             self.switch_btn_start()  # switch button to start
 
-    def stop_recording(self, timeout: bool = False):
+    def stop_recording(self, timeout: bool = False, realtime: bool = False):
         """
         Stop recording
 
         :param timeout: True if stopped due to timeout
+        :param realtime: True if called from realtime callback
         """
         self.plugin.window.core.audio.capture.reset_audio_level()
         self.is_recording = False
@@ -143,7 +183,7 @@ class Simple:
                 return
 
             if self.plugin.window.core.audio.capture.has_frames():
-                if not self.plugin.window.core.audio.capture.has_min_frames():
+                if not self.plugin.window.core.audio.capture.has_min_frames() and not realtime:
                     self.plugin.window.update_status(trans("status.audio.too_short"))
                     self.plugin.window.dispatch(AppEvent(AppEvent.VOICE_CONTROL_STOPPED))  # app event
                     return
@@ -152,6 +192,15 @@ class Simple:
         else:
             self.plugin.window.update_status("")
 
+    def force_stop(self):
+        """Stop recording"""
+        self.is_recording = False
+        self.plugin.window.dispatch(AppEvent(AppEvent.INPUT_VOICE_LISTEN_STOPPED))  # app event
+        self.switch_btn_start()  # switch button to start
+        if self.plugin.window.core.audio.capture.has_source():
+            self.plugin.window.core.audio.capture.stop()  # stop recording
+            return
+
     def on_stop(self):
         """Handle auto-transcribe"""
         path = os.path.join(self.plugin.window.core.config.path, self.plugin.input_file)

pygpt_net/plugin/cmd_files/worker.py

@@ -920,6 +920,9 @@ class Worker(BaseWorker):
         :param context: context data
         :return: extra data
         """
+        # disabled in v2.6.31
+        # reason: do not duplicate context in chat
+        return {}
         cmd = item["cmd"]
         extra = {
             'plugin': "cmd_files",

pygpt_net/provider/api/google/__init__.py

@@ -6,13 +6,14 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.28 20:00:00                  #
+# Updated Date: 2025.08.30 06:00:00                  #
 # ================================================== #
 
 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 +30,7 @@ from .vision import Vision
 from .tools import Tools
 from .audio import Audio
 from .image import Image
-
+from .realtime import Realtime
 
 class ApiGoogle:
     def __init__(self, window=None):
@@ -44,6 +45,7 @@ class ApiGoogle:
         self.tools = Tools(window)
         self.audio = Audio(window)
         self.image = Image(window)
+        self.realtime = Realtime(window)
         self.client: Optional[genai.Client] = None
         self.locked = False
         self.last_client_args: Optional[Dict[str, Any]] = None
@@ -72,12 +74,18 @@ class ApiGoogle:
         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
@@ -94,6 +102,18 @@ 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:
@@ -135,7 +155,11 @@ 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 +230,9 @@ 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 +266,15 @@ 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