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

pygpt_net/CHANGELOG.txt

@@ -1,3 +1,14 @@
+2.6.64 (2025-09-27)
+
+- Added translations to agent headers.
+- Improved presets tabs.
+- Added support for music (Lyria) in both image and video modes (beta).
+
+2.6.63 (2025-09-27)
+
+- Improved agents' workflows.
+- Enhanced the display of agents' steps in the UI.
+
 2.6.62 (2025-09-26)
 
 - Enhanced agent workflow execution.

pygpt_net/__init__.py

@@ -6,15 +6,15 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.26 00:00:00                  #
+# Updated Date: 2025.09.27 00:00:00                  #
 # ================================================== #
 
 __author__ = "Marcin Szczygliński"
 __copyright__ = "Copyright 2025, Marcin Szczygliński"
 __credits__ = ["Marcin Szczygliński"]
 __license__ = "MIT"
-__version__ = "2.6.62"
-__build__ = "2025-09-26"
+__version__ = "2.6.64"
+__build__ = "2025-09-27"
 __maintainer__ = "Marcin Szczygliński"
 __github__ = "https://github.com/szczyglis-dev/py-gpt"
 __report__ = "https://github.com/szczyglis-dev/py-gpt/issues"

pygpt_net/controller/attachment/attachment.py

@@ -579,19 +579,28 @@ class Attachment:
         if not os.path.exists(url):
             return
 
-        if not all:
-            image_ext = ['.png', '.jpg', '.jpeg', '.gif', '.bmp', '.tiff']
-            ext = os.path.splitext(url)[1].lower()
-            if ext not in image_ext:
-                return
+        is_image = False
+        image_ext = ['.png', '.jpg', '.jpeg', '.gif', '.bmp', '.tiff']
+        ext = os.path.splitext(url)[1].lower()
+        if ext in image_ext:
+            is_image = True
+
+        if not all and not is_image:
+            return
+
+        if is_image:
+            title = "attachments.paste.img"
+            status = "painter.capture.manual.captured.success"
+        else:
+            title = "attachments.paste.file"
+            status = "attachments.paste.success"
 
         mode = self.window.core.config.get('mode')
-        title = "Clipboard image"
-        self.window.core.attachments.new(mode, title, url, False)
+        self.window.core.attachments.new(mode, trans(title), url, False)
         self.window.core.attachments.save()
         self.window.controller.attachment.update()
         event = KernelEvent(KernelEvent.STATUS, {
-            'status': trans("painter.capture.manual.captured.success") + ' ' + os.path.basename(url),
+            'status': trans(status) + ' ' + os.path.basename(url),
         })
         self.window.dispatch(event)
 

pygpt_net/controller/camera/camera.py

@@ -67,9 +67,9 @@ class Camera(QObject):
 
         # update label
         if not self.window.core.config.get('vision.capture.auto'):
-            self.window.ui.nodes['video.preview'].label.setText(trans("vision.capture.label"))
+            self.window.ui.nodes['video.preview'].video.setToolTip(trans("vision.capture.label"))
         else:
-            self.window.ui.nodes['video.preview'].label.setText(trans("vision.capture.auto.label"))
+            self.window.ui.nodes['video.preview'].video.setToolTip(trans("vision.capture.auto.label"))
 
     def update(self):
         """Update camera frame"""
@@ -381,7 +381,7 @@ class Camera(QObject):
             {'value': True}
         )
         """
-        self.window.ui.nodes['video.preview'].label.setText(trans("vision.capture.auto.label"))
+        self.window.ui.nodes['video.preview'].video.setToolTip(trans("vision.capture.auto.label"))
 
         if not self.window.core.config.get('vision.capture.enabled'):
             self.enable_capture()
@@ -403,7 +403,7 @@ class Camera(QObject):
             {'value': False}
         )
         """
-        self.window.ui.nodes['video.preview'].label.setText(trans("vision.capture.label"))
+        self.window.ui.nodes['video.preview'].video.setToolTip(trans("vision.capture.label"))
 
     def toggle_auto(self, state: bool):
         """

pygpt_net/controller/lang/custom.py

@@ -68,9 +68,9 @@ class Custom:
 
         # camera capture
         if not self.window.core.config.get('vision.capture.auto'):
-            self.window.ui.nodes['video.preview'].label.setText(trans("vision.capture.label"))
+            self.window.ui.nodes['video.preview'].video.setToolTip(trans("vision.capture.label"))
         else:
-            self.window.ui.nodes['video.preview'].label.setText(trans("vision.capture.auto.label"))
+            self.window.ui.nodes['video.preview'].video.setToolTip(trans("vision.capture.auto.label"))
 
         # files / indexes
         self.window.ui.nodes['output_files'].btn_upload.setText(trans('files.local.upload'))

pygpt_net/controller/presets/editor.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.24 00:00:00                  #
+# Updated Date: 2025.09.27 00:00:00                  #
 # ================================================== #
 
 import datetime
@@ -393,6 +393,9 @@ class Editor:
                         value=extra_options[key].get('default', None),
                     )
 
+            # ensure combo defaults are effectively applied for this tab (only empty values are updated)
+            self._apply_combo_defaults_for_group(option_key, extra_options)
+
     def load_extra_defaults(self):
         """Load extra options defaults for preset editor"""
         if not self.tab_options_idx:
@@ -423,6 +426,8 @@ class Editor:
                             option=extra_options[key],
                             value=value,
                         )
+                # ensure combo defaults are effectively applied for this tab (only empty values are updated)
+                self._apply_combo_defaults_for_group(option_key, extra_options)
 
     def load_extra_defaults_current(self):
         """Load extra options defaults on mode change"""
@@ -479,6 +484,8 @@ class Editor:
                             option=extra_options[key],
                             value=value,
                         )
+                # ensure combo defaults are effectively applied for this tab (only empty values are updated)
+                self._apply_combo_defaults_for_group(option_key, extra_options)
 
     def append_extra_options(self, preset: PresetItem):
         """
@@ -785,6 +792,9 @@ class Editor:
                             value=opt_schema.get('default'),
                         )
 
+            # ensure combo defaults are effectively applied for this tab (only empty values are updated)
+            self._apply_combo_defaults_for_group(config_id, schema_options)
+
         # 4) Recompute mapping fully based on actual tabs and their 'agent_id' properties.
         self._rebuild_tab_index_mapping()
 
@@ -1521,6 +1531,9 @@ class Editor:
                                 value=opt_schema.get('default'),
                             )
 
+                    # ensure combo defaults are effectively applied for this tab (only empty values are updated)
+                    self._apply_combo_defaults_for_group(config_id, schema_options)
+
             # 7) Recompute the index mapping strictly from the QTabWidget
             self._rebuild_tab_index_mapping()
 
@@ -1529,3 +1542,54 @@ class Editor:
 
         finally:
             tabs.setUpdatesEnabled(True)
+
+    # ---------- Helpers for reliable combo defaults in agent extra options ----------
+
+    def _apply_combo_defaults_for_group(self, parent_id: str, schema_options: Dict[str, Any]) -> None:
+        """
+        Ensure that combo-type inputs inside a given UI config group have their default values applied
+        when the current value is empty ("", None or "_"). This avoids the situation where combo boxes
+        remain uninitialized while other field types receive defaults correctly.
+
+        This function never overrides a non-empty value set by the user or loaded from a preset.
+        """
+        if not schema_options:
+            return
+
+        get_value = self.window.controller.config.get_value
+        apply_value = self.window.controller.config.apply_value
+
+        for key, opt_schema in schema_options.items():
+            if not isinstance(opt_schema, dict):
+                continue
+            if opt_schema.get('type') != 'combo':
+                continue
+
+            default_val = opt_schema.get('default', None)
+            if default_val is None:
+                continue
+
+            current_val = get_value(
+                parent_id=parent_id,
+                key=key,
+                option=opt_schema,
+            )
+
+            # Treat "_", "", None as empty and safe to replace with default
+            if current_val in (None, "", "_"):
+                # First try apply_value (standard path)
+                apply_value(
+                    parent_id=parent_id,
+                    key=key,
+                    option=opt_schema,
+                    value=default_val,
+                )
+                # Additionally set directly on widget if accessible to guard against timing of key population
+                try:
+                    widget_group = self.window.ui.config.get(parent_id, {})
+                    widget = widget_group.get(key)
+                    if widget and hasattr(widget, "set_value"):
+                        widget.set_value(default_val)
+                except Exception:
+                    # Silent fallback; apply_value above should already handle most cases
+                    pass

pygpt_net/controller/ui/mode.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.09.22 12:00:00                  #
+# Updated Date: 2025.09.27 15:00:00                  #
 # ================================================== #
 
 from pygpt_net.core.types import (
@@ -58,6 +58,19 @@ class Mode:
         is_completion = mode == MODE_COMPLETION
         is_audio = mode == MODE_AUDIO
 
+        # enable/disable system prompt edit - disable in agents (prompts are defined per agent in presets)
+        if not is_agent_openai and not is_agent_llama:
+            presets_editor.toggle_tab("personalize", True)
+            if 'preset.prompt' in ui_nodes and ui_nodes['preset.prompt'].isReadOnly():
+                ui_nodes['preset.prompt'].setReadOnly(False)
+                ui_nodes['preset.prompt'].setPlaceholderText("")
+        else:
+            presets_editor.toggle_tab("personalize", False)
+            if 'preset.prompt' in ui_nodes and not ui_nodes['preset.prompt'].isReadOnly():
+                ui_nodes['preset.prompt'].setReadOnly(True)
+                ui_nodes['preset.prompt'].setPlaceholderText(trans("toolbox.agent.preset.placeholder"))
+
+        # audio options visibility
         if not is_audio:
             ui_nodes['audio.auto_turn'].setVisible(False)
             ui_nodes["audio.loop"].setVisible(False)
@@ -71,6 +84,7 @@ class Mode:
             else:
                 ctrl.audio.toggle_output_icon(False)
 
+        # presets/assistants visibility
         if not is_assistant:
             ui_nodes['presets.widget'].setVisible(True)
         else:
@@ -81,6 +95,7 @@ class Mode:
         else:
             ui_nodes['env.widget'].setVisible(True)
 
+        # agents/experts/presets label visibility
         show_agents_label = is_agent or is_agent_llama or is_agent_openai
         if show_agents_label:
             ui_nodes['preset.agents.label'].setVisible(True)
@@ -112,6 +127,7 @@ class Mode:
         else:
             ui_nodes['preset.editor.agent_provider_openai'].setVisible(False)
 
+        # prompt editor toolbox visibility
         if is_agent:
             presets_editor.toggle_tab("experts", True)
             ui_nodes['preset.editor.temperature'].setVisible(True)
@@ -145,6 +161,7 @@ class Mode:
             ui_nodes['preset.editor.modes'].setVisible(True)
             ui_tabs['preset.editor.extra'].setTabText(0, trans("preset.prompt"))
 
+        # image options visibility
         if is_image:
             ui_nodes['media.raw'].setVisible(True)
             if ctrl.media.is_video_model():
@@ -198,10 +215,8 @@ class Mode:
         # remote tools icon visibility
         if not is_image and not is_completion:
             self.window.ui.nodes['input'].set_icon_visible("web", True)
-            # ui_nodes['icon.remote_tool.web'].setVisible(True)
         else:
             self.window.ui.nodes['input'].set_icon_visible("web", False)
-            # ui_nodes['icon.remote_tool.web'].setVisible(False)
 
         ui_tabs['input'].setTabVisible(2, is_assistant)
         ui_tabs['input'].setTabVisible(3, (not is_assistant) and (not is_image))

pygpt_net/core/agents/custom/llama_index/runner.py

@@ -1,3 +1,5 @@
+# core/agents/runners/llama_workflow.py
+
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
 # ================================================== #
@@ -6,7 +8,7 @@
 # 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 06:00:00                  #
 # ================================================== #
 
 from __future__ import annotations
@@ -207,37 +209,6 @@ class DynamicFlowWorkflowLI(Workflow):
                 return False
         return False
 
-    def _friendly_map(self) -> Dict[str, str]:
-        return {aid: a.name or aid for aid, a in self.fs.agents.items()}
-
-    def _friendly_map_for_routes(self, route_ids: List[str]) -> Dict[str, Any]:
-        """
-        Build a friendly map for the given route ids:
-        - Always include a human-friendly name.
-        - Include role only if provided in preset options or schema and non-empty.
-        """
-        out: Dict[str, Any] = {}
-        for rid in route_ids or []:
-            a = self.fs.agents.get(rid)
-            name = (a.name if a and a.name else rid)
-            # Prefer preset option, then schema role
-            role_opt = None
-            try:
-                role_opt = self.option_get(rid, "role", None)
-            except Exception:
-                role_opt = None
-            role_schema = getattr(a, "role", None) if a is not None else None
-            role_val = None
-            if isinstance(role_opt, str) and role_opt.strip():
-                role_val = role_opt.strip()
-            elif isinstance(role_schema, str) and role_schema.strip():
-                role_val = role_schema.strip()
-            item = {"name": name}
-            if role_val:
-                item["role"] = role_val
-            out[rid] = item
-        return out
-
     async def _emit(self, ctx: Context, ev: Any):
         if self.dbg.event_echo:
             self.logger.debug(f"[event] emit {ev.__class__.__name__}")
@@ -245,8 +216,8 @@ class DynamicFlowWorkflowLI(Workflow):
 
     async def _emit_agent_text(self, ctx: Context, text: str, agent_name: str = "Agent"):
         """
-        Emit AgentStream(delta=text) robustly. If your env requires extra fields,
-        fall back to extended AgentStream like in your SupervisorWorkflow.
+        Emit AgentStream(delta=text) robustly. If env requires extra fields,
+        fall back to extended AgentStream.
         """
         try:
             if self.dbg.event_echo:
@@ -266,16 +237,11 @@ class DynamicFlowWorkflowLI(Workflow):
             )
 
     async def _emit_header(self, ctx: Context, name: str):
-        if self.dbg.event_echo:
-            self.logger.debug(f"[event] header emit begin name='{name}'")
+        # Lightweight header to ensure agent name is known before tokens.
         await self._emit_agent_text(ctx, "", agent_name=name)
-        # await self._emit_agent_text(ctx, f"\n\n**{name}**\n\n", agent_name=name)
-        if self.dbg.event_echo:
-            self.logger.debug("[event] header emit done")
 
     async def _emit_step_sep(self, ctx: Context, node_id: str):
         try:
-            # Include human-friendly agent name in StepEvent meta for downstream ctx propagation.
             a = self.fs.agents.get(node_id)
             friendly_name = (a.name if a and a.name else node_id)
             await self._emit(
@@ -350,6 +316,9 @@ class DynamicFlowWorkflowLI(Workflow):
             return user_msg, [], "no-mem:last_output"
 
     async def _update_memory_after_step(self, node_id: str, user_msg_text: str, display_text: str):
+        """
+        Update per-node memory after a step, storing baton user message and assistant output.
+        """
         mem_id = self.g.agent_to_memory.get(node_id)
         mem_state = self.mem.get(mem_id) if mem_id else None
         if not mem_state:
@@ -374,7 +343,7 @@ class DynamicFlowWorkflowLI(Workflow):
     # ============== Workflow steps ==============
 
     def run(self, query: str, ctx: Optional[Context] = None, memory: Any = None, verbose: bool = False, on_stop=None):
-        """Entry point used by your LlamaWorkflow runner."""
+        """Entry point used by LlamaWorkflow runner."""
         self._on_stop = on_stop
 
         # Build initial chat once
@@ -444,8 +413,9 @@ class DynamicFlowWorkflowLI(Workflow):
             return FlowTickEvent() if self._current_ids else FlowStopEvent(final_answer=self._last_plain_output or "")
 
         node: AgentNode = self.fs.agents[current_id]
-        if self._steps > 1:
-            await self._emit_step_sep(ctx, current_id)
+
+        # IMPORTANT: emit StepEvent also for the very first agent step.
+        await self._emit_step_sep(ctx, current_id)
         await self._emit_header(ctx, node.name or current_id)
 
         # Resolve runtime + per-node LLM/tools
@@ -474,11 +444,10 @@ class DynamicFlowWorkflowLI(Workflow):
                 f"user='{ellipsize(user_msg_text, self.dbg.preview_chars)}'"
             )
 
-        # Prepare friendly map with optional roles for this node's allowed routes
+        # Build agent
         allowed_routes_now = list(node.outputs or [])
-        friendly_map = self._friendly_map_for_routes(allowed_routes_now)
+        friendly_map = {rid: self.fs.agents.get(rid).name or rid for rid in allowed_routes_now if rid in self.fs.agents}
 
-        # Build agent (chat_history/max_iterations in ctor – best practice)
         built = self.factory.build(
             node=node,
             node_runtime=node_rt,
@@ -528,9 +497,6 @@ class DynamicFlowWorkflowLI(Workflow):
             display_text = decision.content or ""
             if display_text:
                 await self._emit_agent_text(ctx, display_text, agent_name=(node.name or current_id))
-            if self.dbg.log_memory_dump:
-                self.logger.debug(f"[mem.prep] node={current_id} save user='{ellipsize(user_msg_text, self.dbg.preview_chars)}' "
-                                  f"assist='{ellipsize(display_text, self.dbg.preview_chars)}'")
             await self._update_memory_after_step(current_id, user_msg_text, display_text)
             next_id = decision.route if decision.valid else (allowed_routes[0] if allowed_routes else None)
             if self.dbg.log_routes:
@@ -541,9 +507,6 @@ class DynamicFlowWorkflowLI(Workflow):
             display_text = raw_text_clean or ""
             if display_text:
                 await self._emit_agent_text(ctx, display_text, agent_name=(node.name or current_id))
-            if self.dbg.log_memory_dump:
-                self.logger.debug(f"[mem.prep] node={current_id} save user='{ellipsize(user_msg_text, self.dbg.preview_chars)}' "
-                                  f"assist='{ellipsize(display_text, self.dbg.preview_chars)}'")
             await self._update_memory_after_step(current_id, user_msg_text, display_text)
             outs = self.g.get_next(current_id)
             next_id = outs[0] if outs else self.g.first_connected_end(current_id)