pygpt-net 2.7.5__py3-none-any.whl → 2.7.7__py3-none-any.whl

pygpt_net/provider/api/google/utils.py

@@ -0,0 +1,185 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+# ================================================== #
+# This file is a part of PYGPT package               #
+# Website: https://pygpt.net                         #
+# GitHub:  https://github.com/szczyglis-dev/py-gpt   #
+# MIT License                                        #
+# Created By  : Marcin Szczygliński                  #
+# Updated Date: 2026.01.05 20:00:00                  #
+# ================================================== #
+
+from typing import Any, Optional
+
+
+def safe_get(obj: Any, path: str) -> Any:
+    """
+    Dot-path getter for dicts and objects.
+
+    :param obj: Source object or dict
+    :param path: Dot-separated path, e.g. 'a.b.0.c'
+    :return: Value at path or None
+    """
+    cur = obj
+    for seg in path.split("."):
+        if cur is None:
+            return None
+        if isinstance(cur, dict):
+            cur = cur.get(seg)
+        else:
+            if seg.isdigit() and isinstance(cur, (list, tuple)):
+                idx = int(seg)
+                if 0 <= idx < len(cur):
+                    cur = cur[idx]
+                else:
+                    return None
+            else:
+                cur = getattr(cur, seg, None)
+    return cur
+
+
+def as_int(val: Any) -> Optional[int]:
+    """
+    Coerce to int if possible, else None.
+
+    :param val: Input value
+    :return: int or None
+    """
+    if val is None:
+        return None
+    try:
+        return int(val)
+    except Exception:
+        try:
+            return int(float(val))
+        except Exception:
+            return None
+
+def capture_google_usage(state, um_obj: Any):
+    """
+    Extract usage for Google python-genai; prefer total - prompt to include reasoning.
+
+    :param state: Chat state
+    :param um_obj: Usage metadata object/dict
+    """
+    if not um_obj:
+        return
+    state.usage_vendor = "google"
+    prompt = (
+        as_int(safe_get(um_obj, "prompt_token_count")) or
+        as_int(safe_get(um_obj, "prompt_tokens")) or
+        as_int(safe_get(um_obj, "input_tokens"))
+    )
+    total = (
+        as_int(safe_get(um_obj, "total_token_count")) or
+        as_int(safe_get(um_obj, "total_tokens"))
+    )
+    candidates = (
+        as_int(safe_get(um_obj, "candidates_token_count")) or
+        as_int(safe_get(um_obj, "output_tokens"))
+    )
+    reasoning = (
+        as_int(safe_get(um_obj, "candidates_reasoning_token_count")) or
+        as_int(safe_get(um_obj, "reasoning_tokens")) or 0
+    )
+    if total is not None and prompt is not None:
+        out_total = max(0, total - prompt)
+    else:
+        out_total = candidates
+    state.usage_payload = {"in": prompt, "out": out_total, "reasoning": reasoning or 0, "total": total}
+
+
+def collect_google_citations(ctx, state, chunk: Any):
+    """
+    Collect web citations (URLs) from Google GenAI stream.
+
+    :param ctx: Chat context
+    :param state: Chat state
+    :param chunk: Incoming streaming chunk
+    """
+    try:
+        cands = getattr(chunk, "candidates", None) or []
+    except Exception:
+        cands = []
+
+    if not isinstance(state.citations, list):
+        state.citations = []
+
+    def _add_url(url: Optional[str]):
+        if not url or not isinstance(url, str):
+            return
+        url = url.strip()
+        if not (url.startswith("http://") or url.startswith("https://")):
+            return
+        if ctx.urls is None:
+            ctx.urls = []
+        if url not in state.citations:
+            state.citations.append(url)
+        if url not in ctx.urls:
+            ctx.urls.append(url)
+
+    for cand in cands:
+        gm = safe_get(cand, "grounding_metadata") or safe_get(cand, "groundingMetadata")
+        if gm:
+            atts = safe_get(gm, "grounding_attributions") or safe_get(gm, "groundingAttributions") or []
+            try:
+                for att in atts or []:
+                    for path in (
+                        "web.uri",
+                        "web.url",
+                        "source.web.uri",
+                        "source.web.url",
+                        "source.uri",
+                        "source.url",
+                        "uri",
+                        "url",
+                    ):
+                        _add_url(safe_get(att, path))
+            except Exception:
+                pass
+            for path in (
+                "search_entry_point.uri",
+                "search_entry_point.url",
+                "searchEntryPoint.uri",
+                "searchEntryPoint.url",
+                "search_entry_point.rendered_content_uri",
+                "searchEntryPoint.rendered_content_uri",
+            ):
+                _add_url(safe_get(gm, path))
+
+        cm = safe_get(cand, "citation_metadata") or safe_get(cand, "citationMetadata")
+        if cm:
+            cit_arrays = (
+                safe_get(cm, "citation_sources") or
+                safe_get(cm, "citationSources") or
+                safe_get(cm, "citations") or []
+            )
+            try:
+                for cit in cit_arrays or []:
+                    for path in ("uri", "url", "source.uri", "source.url", "web.uri", "web.url"):
+                        _add_url(safe_get(cit, path))
+            except Exception:
+                pass
+
+        try:
+            parts = safe_get(cand, "content.parts") or []
+            for p in parts:
+                pcm = safe_get(p, "citation_metadata") or safe_get(p, "citationMetadata")
+                if pcm:
+                    arr = (
+                        safe_get(pcm, "citation_sources") or
+                        safe_get(pcm, "citationSources") or
+                        safe_get(pcm, "citations") or []
+                    )
+                    for cit in arr or []:
+                        for path in ("uri", "url", "source.uri", "source.url", "web.uri", "web.url"):
+                            _add_url(safe_get(cit, path))
+                gpa = safe_get(p, "grounding_attributions") or safe_get(p, "groundingAttributions") or []
+                for att in gpa or []:
+                    for path in ("web.uri", "web.url", "source.web.uri", "source.web.url", "uri", "url"):
+                        _add_url(safe_get(att, path))
+        except Exception:
+            pass
+
+    if state.citations and (ctx.urls is None or not ctx.urls):
+        ctx.urls = list(state.citations)

pygpt_net/{controller/chat/handler/langchain_stream.py → provider/api/langchain/stream.py}

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.05 00:00:00                  #
+# Updated Date: 2026.01.05 20:00:00                  #
 # ================================================== #
 
 from typing import Optional

pygpt_net/{controller/chat/handler/llamaindex_stream.py → provider/api/llama_index/stream.py}

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.05 00:00:00                  #
+# Updated Date: 2026.01.05 20:00:00                  #
 # ================================================== #
 
 from typing import Optional

pygpt_net/provider/api/openai/__init__.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.12.28 00:00:00                  #
+# Updated Date: 2026.01.03 17:00:00                  #
 # ================================================== #
 
 from openai import OpenAI
@@ -22,6 +22,7 @@ from pygpt_net.core.types import (
     MODE_COMPUTER,
 )
 from pygpt_net.core.bridge.context import BridgeContext
+from pygpt_net.core.types.chunk import ChunkType
 from pygpt_net.item.model import ModelItem
 
 from .audio import Audio
@@ -116,10 +117,12 @@ class ApiOpenAI:
         ctx = context.ctx
         ai_name = ctx.output_name
         thread_id = ctx.thread  # from ctx
+        ctx.chunk_type = ChunkType.API_CHAT  # default: ChatCompletion API
 
         # --- Responses API ----
         use_responses_api = self.responses.is_enabled(model, mode, parent_mode, is_expert_call, preset)
-        ctx.use_responses_api = use_responses_api  # set in context
+        if use_responses_api:
+            ctx.chunk_type = ChunkType.API_CHAT_RESPONSES # Responses API
 
         fixtures = self.window.controller.debug.fixtures
 
@@ -137,13 +140,14 @@ class ApiOpenAI:
 
         # completion
         if mode == MODE_COMPLETION:
+            ctx.chunk_type = ChunkType.API_COMPLETION
             response = self.completion.send(
                 context=context,
                 extra=extra,
             )
             used_tokens = self.completion.get_used_tokens()
 
-        # chat, audio (OpenAI) | research (Perplexity)
+        # chat, audio (OpenAI) | research (deep research, Perplexity)
         elif mode in [
             MODE_CHAT,
             MODE_AUDIO,

pygpt_net/provider/api/openai/image.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.12.31 16:00:00                  #
+# Updated Date: 2026.01.05 20:00:00                  #
 # ================================================== #
 
 import base64
@@ -177,7 +177,7 @@ class ImageWorker(QRunnable):
 
     def _is_gpt_image_model(self, model_id: Optional[str] = None) -> bool:
         mid = (model_id or self.model or "").lower()
-        return mid.startswith("gpt-image-1")
+        return mid.startswith("gpt-image-1") or mid.startswith("chatgpt-image")
 
     def _is_dalle2(self, model_id: Optional[str] = None) -> bool:
         mid = (model_id or self.model or "").lower()

pygpt_net/{controller/chat/handler/openai_stream.py → provider/api/openai/stream.py}

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.12.26 00:00:00                  #
+# Updated Date: 2026.01.05 20:00:00                  #
 # ================================================== #
 
 import base64

pygpt_net/provider/api/openai/utils.py

@@ -6,11 +6,11 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2024.11.26 19:00:00                  #
+# Updated Date: 2026.01.05 20:00:00                  #
 # ================================================== #
 
 import re
-
+from typing import Optional, Any
 
 def sanitize_name(name: str) -> str:
     """
@@ -24,4 +24,70 @@ def sanitize_name(name: str) -> str:
     # allowed characters: a-z, A-Z, 0-9, _, and -
     name = name.strip().lower()
     sanitized_name = re.sub(r'[^a-z0-9_-]', '_', name)
-    return sanitized_name[:64]  # limit to 64 characters
+    return sanitized_name[:64]  # limit to 64 characters
+
+
+def capture_openai_usage(state, u_obj: Any):
+    """
+    Extract usage for OpenAI/xAI-compatible chunks.
+
+    :param state: Chat state
+    :param u_obj: Usage object/dict
+    """
+    if not u_obj:
+        return
+    state.usage_vendor = "openai"
+    in_tok = as_int(safe_get(u_obj, "input_tokens")) or as_int(safe_get(u_obj, "prompt_tokens"))
+    out_tok = as_int(safe_get(u_obj, "output_tokens")) or as_int(safe_get(u_obj, "completion_tokens"))
+    total = as_int(safe_get(u_obj, "total_tokens"))
+    reasoning = (
+        as_int(safe_get(u_obj, "output_tokens_details.reasoning_tokens")) or
+        as_int(safe_get(u_obj, "completion_tokens_details.reasoning_tokens")) or
+        as_int(safe_get(u_obj, "reasoning_tokens")) or
+        0
+    )
+    out_with_reason = (out_tok or 0) + (reasoning or 0)
+    state.usage_payload = {"in": in_tok, "out": out_with_reason, "reasoning": reasoning or 0, "total": total}
+
+def safe_get(obj: Any, path: str) -> Any:
+    """
+    Dot-path getter for dicts and objects.
+
+    :param obj: Source object or dict
+    :param path: Dot-separated path, e.g. 'a.b.0.c'
+    :return: Value at path or None
+    """
+    cur = obj
+    for seg in path.split("."):
+        if cur is None:
+            return None
+        if isinstance(cur, dict):
+            cur = cur.get(seg)
+        else:
+            if seg.isdigit() and isinstance(cur, (list, tuple)):
+                idx = int(seg)
+                if 0 <= idx < len(cur):
+                    cur = cur[idx]
+                else:
+                    return None
+            else:
+                cur = getattr(cur, seg, None)
+    return cur
+
+
+def as_int(val: Any) -> Optional[int]:
+    """
+    Coerce to int if possible, else None.
+
+    :param val: Input value
+    :return: int or None
+    """
+    if val is None:
+        return None
+    try:
+        return int(val)
+    except Exception:
+        try:
+            return int(float(val))
+        except Exception:
+            return None

pygpt_net/provider/api/x_ai/__init__.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.15 01:00:00                  #
+# Updated Date: 2026.01.04 19:00:00                  #
 # ================================================== #
 
 from typing import Optional, Dict, Any
@@ -23,6 +23,7 @@ from pygpt_net.core.types import (
     MODE_RESEARCH,
 )
 from pygpt_net.core.bridge.context import BridgeContext
+from pygpt_net.core.types.chunk import ChunkType
 from pygpt_net.item.model import ModelItem
 
 import xai_sdk
@@ -32,7 +33,8 @@ from .vision import Vision
 from .tools import Tools
 from .audio import Audio
 from .image import Image
-from .remote import Remote
+from .remote_tools import Remote
+from .responses import Responses
 
 
 class ApiXAI:
@@ -48,7 +50,8 @@ class ApiXAI:
         self.tools = Tools(window)
         self.audio = Audio(window)
         self.image = Image(window)
-        self.remote = Remote(window)  # Live Search builder
+        self.remote = Remote(window)
+        self.responses = Responses(window)
         self.client: Optional[xai_sdk.Client] = None
         self.locked = False
         self.last_client_args: Optional[Dict[str, Any]] = None
@@ -100,8 +103,9 @@ class ApiXAI:
         """
         Make an API call to xAI.
 
-        Supports chat (stream/non-stream), images (via REST),
-        and function-calling. Audio is not available in public xAI SDK at this time.
+        Uses old API and Chat Responses (stateful) via xai_sdk:
+        - Streaming: chat.stream() (tuples of (response, chunk))
+        - Non-stream: chat.sample()
 
         :param context: BridgeContext
         :param extra: Extra params (not used)
@@ -109,22 +113,33 @@ class ApiXAI:
         :return: True on success, False on error
         """
         mode = context.mode
-        model = context.model
         stream = context.stream
         ctx = context.ctx
         ai_name = (ctx.output_name if ctx else "assistant")
-
-        # No Responses API in xAI SDK
-        if ctx:
-            ctx.use_responses_api = False
-
+        model = context.model  # model instance (item, not id)
         used_tokens = 0
         response = None
+        ctx.chunk_type = ChunkType.XAI_SDK
+        
+        use_responses_api = True
+        if model and model.id.startswith("grok-3"):
+            use_responses_api = False  # use old API
+
+        if mode in (
+                MODE_COMPLETION,
+                MODE_CHAT,
+                MODE_AUDIO,
+                MODE_RESEARCH
+        ):
+            # Audio TTS is not exposed via public SDK; treat MODE_AUDIO as chat input.
+            # NOTE: for grok-3 use Chat completions, for > grok-4 use Chat responses
+            if use_responses_api:
+                response = self.responses.send(context=context, extra=extra)  # responses
+                used_tokens = self.responses.get_used_tokens()
+            else:
+                response = self.chat.send(context=context, extra=extra)  # completions
+                used_tokens = self.chat.get_used_tokens()
 
-        if mode in (MODE_COMPLETION, MODE_CHAT, MODE_AUDIO, MODE_RESEARCH):
-            # There is no public realtime audio in SDK; treat MODE_AUDIO as chat (TTS not supported).
-            response = self.chat.send(context=context, extra=extra)
-            used_tokens = self.chat.get_used_tokens()
             if ctx:
                 self.vision.append_images(ctx)
 
@@ -150,7 +165,10 @@ class ApiXAI:
 
         if ctx:
             ctx.ai_name = ai_name
-            self.chat.unpack_response(context.mode, response, ctx)
+            if use_responses_api:
+                self.responses.unpack_response(context.mode, response, ctx)
+            else:
+                self.chat.unpack_response(context.mode, response, ctx)
             try:
                 for tc in getattr(ctx, "tool_calls", []) or []:
                     fn = tc.get("function") or {}
@@ -174,6 +192,88 @@ class ApiXAI:
 
         If context.request is set, makes a full call() instead (for consistency).
 
+        :param context: BridgeContext
+        :param extra: Extra params (not used)
+        :return: Output text or "" on error
+        """
+        model = context.model or self.window.core.models.from_defaults()
+        if model and model.id.startswith("grok-3"):
+            return self.quick_call_old(context, extra)  # grok-3 uses old path
+
+        if context.request:
+            context.stream = False
+            context.mode = MODE_CHAT
+            self.locked = True
+            self.call(context, extra)
+            self.locked = False
+            return context.ctx.output
+
+        self.locked = True
+        try:
+            ctx = context.ctx
+            prompt = context.prompt
+            system_prompt = context.system_prompt
+            history = context.history
+            functions = context.external_functions
+            attachments = context.attachments
+            multimodal_ctx = context.multimodal_ctx
+
+            # Prepare client-side tools for SDK (no server-side tools in quick_call)
+            client_tools = self.tools.prepare_sdk_tools(functions)
+
+            client = self.get_client(MODE_CHAT, model)
+            # store_messages: false for quick, and false if images present (SDK guidance)
+            store_messages = False
+            prev_id = None
+
+            # Create chat session
+            include = []
+            chat = client.chat.create(
+                model=model.id,
+                tools=(client_tools if client_tools else None),
+                include=(include if include else None),
+                store_messages=store_messages,
+                previous_response_id=prev_id,
+            )
+
+            # Append history if enabled and no previous_response_id is used
+            self.responses.append_history_sdk(
+                chat=chat,
+                system_prompt=system_prompt,
+                model=model,
+                history=history,
+            )
+
+            # Append current prompt with optional images
+            self.responses.append_current_user_sdk(
+                chat=chat,
+                prompt=prompt,
+                attachments=attachments,
+                multimodal_ctx=multimodal_ctx,
+            )
+
+            resp = chat.sample()
+            # Extract client-side tool calls if any (leave server-side out)
+            out = getattr(resp, "content", "") or ""
+            if ctx:
+                self.responses.quick_collect_response_id(resp, ctx)
+            return out.strip()
+        except Exception as e:
+            self.window.core.debug.log(e)
+            return ""
+        finally:
+            self.locked = False
+
+    def quick_call_old(
+            self,
+            context: BridgeContext,
+            extra: dict = None
+    ) -> str:
+        """
+        Quick non-streaming xAI chat call and return output text.
+
+        If context.request is set, makes a full call() instead (for consistency).
+
         :param context: BridgeContext
         :param extra: Extra params (not used)
         :return: Output text or "" on error
@@ -201,7 +301,7 @@ class ApiXAI:
             # If tools are present, prefer non-streaming HTTP Chat Completions path to extract tool calls reliably.
             # Otherwise use native SDK chat.sample().
             if tools:
-                out, calls, citations, usage  = self.chat.call_http_nonstream(
+                out, calls, citations, usage = self.chat.call_http_nonstream(
                     model=model.id,
                     prompt=prompt,
                     system_prompt=system_prompt,