pygpt-net 2.6.62__py3-none-any.whl → 2.6.64__py3-none-any.whl

pygpt_net/provider/agents/openai/agent_with_experts_feedback.py

@@ -342,13 +342,13 @@ class Agent(BaseAgent):
                         "type": "bool",
                         "label": trans("agent.option.tools.local"),
                         "description": trans("agent.option.tools.local.desc"),
-                        "default": False,
+                        "default": True,
                     },
                     "allow_remote_tools": {
                         "type": "bool",
                         "label": trans("agent.option.tools.remote"),
                         "description": trans("agent.option.tools.remote.desc"),
-                        "default": False,
+                        "default": True,
                     },
                 }
             },
@@ -371,13 +371,13 @@ class Agent(BaseAgent):
                         "type": "bool",
                         "label": trans("agent.option.tools.local"),
                         "description": trans("agent.option.tools.local.desc"),
-                        "default": False,
+                        "default": True,
                     },
                     "allow_remote_tools": {
                         "type": "bool",
                         "label": trans("agent.option.tools.remote"),
                         "description": trans("agent.option.tools.remote.desc"),
-                        "default": False,
+                        "default": True,
                     },
                 }
             },

pygpt_net/provider/agents/openai/agent_with_feedback.py

@@ -342,13 +342,13 @@ class Agent(BaseAgent):
                         "type": "bool",
                         "label": trans("agent.option.tools.local"),
                         "description": trans("agent.option.tools.local.desc"),
-                        "default": False,
+                        "default": True,
                     },
                     "allow_remote_tools": {
                         "type": "bool",
                         "label": trans("agent.option.tools.remote"),
                         "description": trans("agent.option.tools.remote.desc"),
-                        "default": False,
+                        "default": True,
                     },
                 }
             },
@@ -371,13 +371,13 @@ class Agent(BaseAgent):
                         "type": "bool",
                         "label": trans("agent.option.tools.local"),
                         "description": trans("agent.option.tools.local.desc"),
-                        "default": False,
+                        "default": True,
                     },
                     "allow_remote_tools": {
                         "type": "bool",
                         "label": trans("agent.option.tools.remote"),
                         "description": trans("agent.option.tools.remote.desc"),
-                        "default": False,
+                        "default": True,
                     },
                 }
             },

pygpt_net/provider/agents/openai/evolve.py

@@ -330,7 +330,6 @@ class Agent(BaseAgent):
                 choose_query = self.make_choose_query(results)
                 choose_items.append(choose_query)
 
-                ctx.set_agent_name(chooser.name)
                 chooser_result = await Runner.run(chooser, choose_items)
                 result: ChooseFeedback = chooser_result.final_output
                 choose = result.answer_number
@@ -348,7 +347,6 @@ class Agent(BaseAgent):
                     bridge.on_stop(ctx)
                     break
 
-                ctx.set_agent_name(evaluator.name)
                 evaluator_result = await Runner.run(evaluator, input_items)
                 result: EvaluationFeedback = evaluator_result.final_output
 
@@ -443,7 +441,6 @@ class Agent(BaseAgent):
                 window.core.api.openai.responses.unpack_agent_response(results[choose], ctx)
                 input_items = results[choose].to_input_list()
 
-                ctx.set_agent_name(evaluator.name)
                 evaluator_result = await Runner.run(evaluator, input_items)
                 result: EvaluationFeedback = evaluator_result.final_output
 
@@ -528,13 +525,13 @@ class Agent(BaseAgent):
                         "type": "bool",
                         "label": trans("agent.option.tools.local"),
                         "description": trans("agent.option.tools.local.desc"),
-                        "default": False,
+                        "default": True,
                     },
                     "allow_remote_tools": {
                         "type": "bool",
                         "label": trans("agent.option.tools.remote"),
                         "description": trans("agent.option.tools.remote.desc"),
-                        "default": False,
+                        "default": True,
                     },
                 }
             },
@@ -557,13 +554,13 @@ class Agent(BaseAgent):
                         "type": "bool",
                         "label": trans("agent.option.tools.local"),
                         "description": trans("agent.option.tools.local.desc"),
-                        "default": False,
+                        "default": True,
                     },
                     "allow_remote_tools": {
                         "type": "bool",
                         "label": trans("agent.option.tools.remote"),
                         "description": trans("agent.option.tools.remote.desc"),
-                        "default": False,
+                        "default": True,
                     },
                 }
             },
@@ -586,13 +583,13 @@ class Agent(BaseAgent):
                         "type": "bool",
                         "label": trans("agent.option.tools.local"),
                         "description": trans("agent.option.tools.local.desc"),
-                        "default": False,
+                        "default": True,
                     },
                     "allow_remote_tools": {
                         "type": "bool",
                         "label": trans("agent.option.tools.remote"),
                         "description": trans("agent.option.tools.remote.desc"),
-                        "default": False,
+                        "default": True,
                     },
                 }
             },

pygpt_net/provider/agents/openai/supervisor.py

@@ -6,12 +6,13 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.26 17:00:00                  #
+# Updated Date: 2025.09.27 20:25:00                  #
 # ================================================== #
-
+import base64
 import json
 import re
-from typing import Dict, Any, Tuple, Optional
+import io
+from typing import Dict, Any, Tuple, Optional, Callable
 
 from agents import (
     Agent as OpenAIAgent,
@@ -38,6 +39,16 @@ from pygpt_net.utils import trans
 
 from ..base import BaseAgent
 
+# OpenAI response event types (used by StreamHandler)
+from openai.types.responses import (
+    ResponseTextDeltaEvent,
+    ResponseCreatedEvent,
+    ResponseCodeInterpreterCallCodeDeltaEvent,
+    ResponseOutputItemAddedEvent,
+    ResponseCompletedEvent,
+    ResponseOutputItemDoneEvent,
+)
+
 JSON_RE = re.compile(r"\{[\s\S]*\}$", re.MULTILINE)
 
 SUPERVISOR_PROMPT = """
@@ -62,6 +73,200 @@ You are the “Worker”. You execute Supervisor instructions strictly, using yo
 Respond in the user's language.
 """
 
+
+class SupervisorStreamHandler(StreamHandler):
+    """
+    Stream handler that filters JSON from Supervisor output during streaming.
+    - Pass-through normal text.
+    - Suppress raw JSON (both ```json fenced and bare {...}).
+    - When JSON finishes, parse and emit only the human-friendly text via `json_to_text`.
+    """
+    def __init__(
+        self,
+        window,
+        bridge: ConnectionContext = None,
+        message: str = None,
+        json_to_text: Optional[Callable[[dict], str]] = None,
+    ):
+        super().__init__(window, bridge, message)
+        self.json_to_text = json_to_text or (lambda d: json.dumps(d, ensure_ascii=False))
+        self._json_fenced = False
+        self._json_buf = io.StringIO()
+        self._json_in_braces = False
+        self._brace_depth = 0
+        self._in_string = False
+        self._escape = False
+
+    def _emit_text(self, ctx: CtxItem, text: str, flush: bool, buffer: bool):
+        if not text:
+            return
+        self._emit(ctx, text, flush, buffer)
+
+    def _flush_json(self, ctx: CtxItem, flush: bool, buffer: bool):
+        """
+        Parse collected JSON and emit only formatted text; reset state.
+        """
+        raw_json = self._json_buf.getvalue().strip()
+        self._json_buf = io.StringIO()
+        self._json_fenced = False
+        self._json_in_braces = False
+        self._brace_depth = 0
+        self._in_string = False
+        self._escape = False
+
+        if not raw_json:
+            return
+        try:
+            data = json.loads(raw_json)
+            out = self.json_to_text(data) or ""
+        except Exception:
+            # Fallback: if parsing failed, do not leak JSON; stay silent
+            out = ""
+        if out:
+            self._emit_text(ctx, out, flush, buffer)
+
+    def _handle_text_delta(self, s: str, ctx: CtxItem, flush: bool, buffer: bool):
+        """
+        Filter JSON while streaming; emit only non-JSON text or parsed JSON text.
+        """
+        i = 0
+        n = len(s)
+        while i < n:
+            # Detect fenced JSON start
+            if not self._json_fenced and not self._json_in_braces and s.startswith("```json", i):
+                # Emit any text before the fence
+                # (there shouldn't be in this branch because we check exact start, but keep safe)
+                i += len("```json")
+                self._json_fenced = True
+                # Skip possible newline after fence
+                if i < n and s[i] == '\n':
+                    i += 1
+                continue
+
+            # Detect fenced JSON end
+            if self._json_fenced and s.startswith("```", i):
+                # Flush JSON collected so far
+                self._flush_json(ctx, flush, buffer)
+                i += len("```")
+                # Optional newline after closing fence
+                if i < n and s[i] == '\n':
+                    i += 1
+                continue
+
+            # While inside fenced JSON -> buffer and continue
+            if self._json_fenced:
+                self._json_buf.write(s[i])
+                i += 1
+                continue
+
+            # Bare JSON detection (naive but effective for supervisor outputs)
+            if not self._json_in_braces and s[i] == "{":
+                self._json_in_braces = True
+                self._brace_depth = 1
+                self._in_string = False
+                self._escape = False
+                self._json_buf.write("{")
+                i += 1
+                continue
+
+            if self._json_in_braces:
+                ch = s[i]
+                # Basic JSON string/escape handling
+                if ch == '"' and not self._escape:
+                    self._in_string = not self._in_string
+                if ch == "\\" and not self._escape:
+                    self._escape = True
+                else:
+                    self._escape = False
+                if not self._in_string:
+                    if ch == "{":
+                        self._brace_depth += 1
+                    elif ch == "}":
+                        self._brace_depth -= 1
+                self._json_buf.write(ch)
+                i += 1
+                if self._brace_depth == 0:
+                    # JSON closed -> flush parsed text
+                    self._flush_json(ctx, flush, buffer)
+                continue
+
+            # Normal text path
+            # Accumulate until potential fenced start to avoid splitting too often
+            next_fence = s.find("```json", i)
+            next_bare = s.find("{", i)
+            cut = n
+            candidates = [x for x in (next_fence, next_bare) if x != -1]
+            if candidates:
+                cut = min(candidates)
+            chunk = s[i:cut]
+            if chunk:
+                self._emit_text(ctx, chunk, flush, buffer)
+            i = cut if cut != n else n
+
+    def handle(
+        self,
+        event,
+        ctx: CtxItem,
+        flush: bool = True,
+        buffer: bool = True
+    ) -> Tuple[str, str]:
+        """
+        Override StreamHandler.handle to filter JSON in text deltas.
+        For non-text events, fallback to parent handler.
+        """
+        # ReasoningItem path remains the same (parent prints to stdout), keep parent behavior.
+
+        if getattr(event, "type", None) == "raw_response_event":
+            data = event.data
+
+            if isinstance(data, ResponseCreatedEvent):
+                self.response_id = data.response.id
+                return self.buffer, self.response_id
+
+            if isinstance(data, ResponseTextDeltaEvent):
+                # Filter JSON while streaming
+                delta = data.delta or ""
+                # If a code_interpreter block was started previously, render fence first
+                if self.code_block:
+                    self._emit_text(ctx, "\n```\n", flush, buffer)
+                    self.code_block = False
+                self._handle_text_delta(delta, ctx, flush, buffer)
+                return self.buffer, self.response_id
+
+            if isinstance(data, ResponseOutputItemAddedEvent):
+                if data.item.type == "code_interpreter_call":
+                    self.code_block = True
+                    s = "\n\n**Code interpreter**\n```python\n"
+                    self._emit_text(ctx, s, flush, buffer)
+                return self.buffer, self.response_id
+
+            if isinstance(data, ResponseOutputItemDoneEvent):
+                if data.item.type == "image_generation_call":
+                    img_path = self.window.core.image.gen_unique_path(ctx)
+                    image_base64 = data.item.result
+                    image_bytes = base64.b64decode(image_base64)
+                    with open(img_path, "wb") as f:
+                        f.write(image_bytes)
+                    self.window.core.debug.info("[chat] Image generation call found")
+                    ctx.images = [img_path]
+                return self.buffer, self.response_id
+
+            if isinstance(data, ResponseCodeInterpreterCallCodeDeltaEvent):
+                self._emit_text(ctx, data.delta or "", flush, buffer)
+                return self.buffer, self.response_id
+
+            if isinstance(data, ResponseCompletedEvent):
+                # If we are still buffering JSON, flush it now (emit parsed text only)
+                if self._json_fenced or self._json_in_braces:
+                    self._flush_json(ctx, flush, buffer)
+                # Mark finished so parent downloader logic (files) may trigger if needed
+                self.finished = True
+                return self.buffer, self.response_id
+
+        # Handoff / other events: fallback to parent, but it won't print JSON since we already filtered in text deltas
+        return super().handle(event, ctx, flush, buffer)
+
+
 class Agent(BaseAgent):
 
     def __init__(self, *args, **kwargs):
@@ -69,7 +274,7 @@ class Agent(BaseAgent):
         self.id = "openai_agent_supervisor"
         self.type = AGENT_TYPE_OPENAI
         self.mode = AGENT_MODE_OPENAI
-        self.name = "Supervisor + worker"
+        self.name = "Supervisor"  # use clean name in UI headers
 
     def get_agent(self, window, kwargs: Dict[str, Any]):
         """
@@ -81,8 +286,11 @@ class Agent(BaseAgent):
         """
         context = kwargs.get("context", BridgeContext())
         preset = context.preset
-        agent_name = preset.name if preset else "Supervisor"
         model = kwargs.get("model", ModelItem())
+
+        # Enforce a stable, clean display name for the Supervisor regardless of preset name.
+        agent_name = "Supervisor"  # hard-coded UI name
+
         worker_tool = kwargs.get("worker_tool", None)
         kwargs = {
             "name": agent_name,
@@ -184,65 +392,117 @@ class Agent(BaseAgent):
         worker_session_id = f"worker_session_{ctx.meta.id}" if ctx.meta else "worker_session_default"
         worker_session = SQLiteSession(worker_session_id)
 
-        handler = StreamHandler(window, bridge)
-        item_ctx = ctx
+        # Use JSON-filtering handler for Supervisor streaming
+        handler = SupervisorStreamHandler(
+            window,
+            bridge,
+            json_to_text=self.response_from_json,
+        )
+        item_ctx = ctx  # will reassign on splits
+
+        supervisor_display_name = None  # set after agent is created
 
         # tool to run Worker
         @function_tool(name_override="run_worker")
-        async def run_worker(ctx: RunContextWrapper[Any], instruction: str) -> str:
+        async def run_worker(fn_ctx: RunContextWrapper[Any], instruction: str) -> str:
             """
             Run the Worker with an instruction from the Supervisor and return its output.
 
-            :param ctx: Run context wrapper
-            :param instruction: Instruction for the Worker
-            :return: Output from the Worker
+            - Appends the instruction to the current Supervisor block.
+            - Finalizes the Supervisor block and opens a new Worker block.
+            - Runs the Worker and streams its result into the Worker block.
+            - Finalizes the Worker block, then opens a fresh block for the Supervisor to continue.
             """
-            item_ctx.stream = f"\n\n**{trans('agent.name.supervisor')} --> {trans('agent.name.worker')}:** {instruction}\n\n"
+            nonlocal item_ctx, supervisor_display_name
+
+            info = f"\n\n**{trans('agent.name.supervisor')} → {trans('agent.name.worker')}:** {instruction}\n\n"
+            item_ctx.stream = info
             bridge.on_step(item_ctx, True)
-            handler.begin = False
+            handler.to_buffer(info)
+
+            if use_partial_ctx:
+                item_ctx = bridge.on_next_ctx(
+                    ctx=item_ctx,
+                    input="",
+                    output=handler.buffer,  # finalize current Supervisor content
+                    response_id=handler.response_id or "",
+                    stream=True,
+                )
+                handler.new()  # reset handler buffer for next block
+
+            try:
+                item_ctx.set_agent_name(worker.name)
+            except Exception:
+                pass
+
             result = await Runner.run(
                 worker,
                 input=instruction,
                 session=worker_session,
                 max_turns=max_steps,
             )
-            item_ctx.stream = f"\n\n{result.final_output}\n\n"
-            bridge.on_step(item_ctx, False)
-            return str(result.final_output)
+
+            worker_text = str(result.final_output or "")
+            if worker_text:
+                item_ctx.stream = f"{worker_text}\n"
+                bridge.on_step(item_ctx, True)
+
+            if use_partial_ctx:
+                item_ctx = bridge.on_next_ctx(
+                    ctx=item_ctx,
+                    input="",
+                    output=worker_text,  # finalize worker output
+                    response_id="",      # worker has no OpenAI response id here
+                    stream=True,
+                )
+                try:
+                    if supervisor_display_name:
+                        item_ctx.set_agent_name(supervisor_display_name)
+                except Exception:
+                    pass
+
+            return worker_text
 
         agent_kwargs["worker_tool"] = run_worker
         agent = self.get_agent(window, agent_kwargs)
+        supervisor_display_name = agent.name  # "Supervisor"
 
         if not stream:
             item_ctx.set_agent_name(agent.name)
-            result = await Runner.run(
-                agent,
-                **kwargs
-            )
-            final_output, last_response_id = window.core.api.openai.responses.unpack_agent_response(result, ctx)
+            result = await Runner.run(agent, **kwargs)
+            final_output, last_response_id = window.core.api.openai.responses.unpack_agent_response(result, item_ctx)
             response_id = result.last_response_id
             if verbose:
                 print("Final response:", result)
         else:
             item_ctx.set_agent_name(agent.name)
-            result = Runner.run_streamed(
-                agent,
-                **kwargs
-            )
+            result = Runner.run_streamed(agent, **kwargs)
             async for event in result.stream_events():
                 if bridge.stopped():
                     result.cancel()
-                    bridge.on_stop(ctx)
+                    bridge.on_stop(item_ctx)
                     break
-                final_output, response_id = handler.handle(event, ctx)
+                # Write into current item_ctx (it changes when we split)
+                final_output, response_id = handler.handle(event, item_ctx)
 
-        # extract final output from JSON
+        # extract final output from JSON (Supervisor's last block)
         if final_output:
             final_output = self.extract_final_response(final_output)
             if verbose:
                 print("Final output after extraction:", final_output)
 
-        return ctx, final_output, response_id
+        # Properly finalize last block
+        if use_partial_ctx:
+            item_ctx = bridge.on_next_ctx(
+                ctx=item_ctx,
+                input=final_output or "",
+                output=final_output or "",
+                response_id=response_id or "",
+                finish=True,
+                stream=stream,
+            )
+
+        return item_ctx, final_output, response_id
 
     def extract_final_response(self, output: str) -> str:
         """
@@ -257,7 +517,6 @@ class Agent(BaseAgent):
         fence = re.search(r"```json\s*([\s\S]*?)\s*```", output, re.IGNORECASE)
         if fence:
             try:
-                # Try to parse the fenced JSON
                 json_text = fence.group(1).strip()
                 json_response = json.loads(json_text)
                 return self.response_from_json(json_response)
@@ -267,7 +526,6 @@ class Agent(BaseAgent):
         tail = JSON_RE.findall(output)
         for candidate in tail[::-1]:
             try:
-                # Try to parse the JSON from the tail
                 json_response = json.loads(candidate)
                 return self.response_from_json(json_response)
             except Exception:
@@ -275,7 +533,6 @@ class Agent(BaseAgent):
 
         if output.startswith("{") and output.endswith("}"):
             try:
-                # Try to parse the entire output as JSON
                 response = json.loads(output)
                 return self.response_from_json(response)
             except Exception as e:
@@ -350,8 +607,4 @@ class Agent(BaseAgent):
                     },
                 }
             },
-        }
-
-
-
-
+        }

pygpt_net/provider/api/google/__init__.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.17 05:00:00                  #
+# Updated Date: 2025.09.14 00:00:00                  #
 # ================================================== #
 
 import os
@@ -33,6 +33,7 @@ from .audio import Audio
 from .image import Image
 from .realtime import Realtime
 from .video import Video
+from .music import Music
 
 class ApiGoogle:
     def __init__(self, window=None):
@@ -49,6 +50,7 @@ class ApiGoogle:
         self.image = Image(window)
         self.realtime = Realtime(window)
         self.video = Video(window)
+        self.music = Music(window)
         self.client: Optional[genai.Client] = None
         self.locked = False
         self.last_client_args: Optional[Dict[str, Any]] = None
@@ -135,10 +137,14 @@ class ApiGoogle:
                 self.vision.append_images(ctx)
 
         elif mode == MODE_IMAGE:
+            # Route to video / music / image based on selected model.
             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.
+            # Lyria / music models
+            if self.music.is_music_model(model.id if model else ""):
+                return self.music.generate(context=context, extra=extra)   # lyria, etc.
+            # Default: image
+            return self.image.generate(context=context, extra=extra)       # imagen, etc.
 
         elif mode == MODE_ASSISTANT:
             return False  # not implemented for Google

pygpt_net/provider/api/google/image.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.29 20:40:00                  #
+# Updated Date: 2025.09.14 00:00:00                  #
 # ================================================== #
 
 import mimetypes
@@ -45,6 +45,16 @@ class Image:
         :param sync: run synchronously (blocking) if True
         :return: True if started
         """
+        # Music fast-path: delegate to Music flow if a music model is selected (e.g., Lyria).
+        # This keeps image flow unchanged while enabling music in the same "image" mode.
+        try:
+            model_id = (context.model.id if context and context.model else "") or ""
+            if self.window and hasattr(self.window.core.api.google, "music"):
+                if self.window.core.api.google.music.is_music_model(model_id):
+                    return self.window.core.api.google.music.generate(context=context, extra=extra, sync=sync)
+        except Exception:
+            pass
+
         extra = extra or {}
         ctx = context.ctx or CtxItem()
         model = context.model