pygpt-net 2.6.20__py3-none-any.whl → 2.6.21__py3-none-any.whl

pygpt_net/controller/kernel/reply.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.07.01 01:00:00                  #
+# Updated Date: 2025.08.23 15:00:00                  #
 # ================================================== #
 
 import json
@@ -20,7 +20,7 @@ from pygpt_net.item.ctx import CtxItem
 class Reply:
     def __init__(self, window=None):
         """
-        Reply handler
+        Reply handler (response from plugins, tools, etc.)
 
         :param window: Window instance
         """
@@ -42,28 +42,33 @@ class Reply:
         :param extra: extra data
         :return: list of results
         """
-        flush = False
-        if "flush" in extra and extra["flush"]:
-            flush = True
+        flush = extra.get("flush", False) if isinstance(extra, dict) else False
         ctx = context.ctx
-        if ctx is not None:
-            self.run_post_response(ctx, extra)
-            self.last_result = ctx.results
-            if ctx.agent_call:
-                return ctx.results # abort if called in agent and return here, TODO: check if needed!!!!!
-            self.window.core.debug.info("Reply...")
-            if self.window.core.debug.enabled() and self.is_log():
-                self.window.core.debug.debug("CTX REPLY: " + str(ctx))
-            if ctx.reply:
-                if self.reply_idx >= ctx.pid:  # skip if reply already sent for this context
-                    # >>> this prevents multiple replies from the same ctx item <<<
-                    return []
-                self.reply_idx = ctx.pid 
-                self.append(ctx)
-            if flush or self.window.controller.kernel.async_allowed(ctx):
-                self.flush()
-            return ctx.results
-        return []
+        if ctx is None:
+            return []
+
+        core = self.window.core
+        self.last_result = ctx.results
+        self.on_post_response(ctx, extra)
+
+        if ctx.agent_call:
+            # TODO: clear() here?
+            return ctx.results # abort if called by agent, TODO: check if needed!!!!!
+
+        core.debug.info("Reply...")
+        if core.debug.enabled() and self.is_log():
+            core.debug.debug("CTX REPLY: " + str(ctx))
+
+        if ctx.reply:
+            if self.reply_idx >= ctx.pid:  # prevent multiple replies per ctx
+                return []
+            self.reply_idx = ctx.pid
+            self.append(ctx)
+
+        if flush or self.window.controller.kernel.async_allowed(ctx):
+            self.flush()
+
+        return ctx.results
 
     def append(self, ctx: CtxItem):
         """
@@ -73,76 +78,63 @@ class Reply:
         """
         self.window.core.debug.info("Reply stack (add)...")
         self.reply_stack.append(ctx.results)
-        ctx.results = []  # clear results
         self.reply_ctx = ctx
+        self.reply_ctx.results = []  # clear results
 
     def flush(self):
         """Flush reply stack"""
         if self.reply_ctx is None or len(self.reply_stack) == 0:
             return
 
-        self.window.core.debug.info("Reply stack (flush)...")
+        core = self.window.core
+        dispatch = self.window.dispatch
+        core.debug.info("Reply stack (flush)...")
+
         results = []
         for responses in self.reply_stack:
             for result in responses:
                 results.append(result)
 
         self.window.update_status("")  # clear status
-        if self.reply_ctx.internal:
-            if self.window.controller.agent.legacy.enabled():
-                self.window.controller.agent.legacy.add_run()
-                self.window.controller.agent.legacy.update()
+        self.window.controller.agent.on_reply(self.reply_ctx)  # handle reply in agent
 
         # prepare data to send as reply
         tool_data = json.dumps(results)
         if (len(self.reply_stack) < 2
                 and self.reply_ctx.extra_ctx
-                and self.window.core.config.get("ctx.use_extra")):
+                and core.config.get("ctx.use_extra")):
             tool_data = self.reply_ctx.extra_ctx  # if extra content is set, use it as data to send
 
-        prev_ctx = self.window.core.ctx.as_previous(self.reply_ctx)  # copy result to previous ctx and clear current ctx
-        self.window.core.ctx.update_item(self.reply_ctx)  # update context in db
+        prev_ctx = core.ctx.as_previous(self.reply_ctx)  # copy result to previous ctx and clear current ctx
+        core.ctx.update_item(self.reply_ctx)  # update context in db
         self.window.update_status('...')
 
-        # if response from sub call, from experts
-        parent_id = None
-        if self.reply_ctx.sub_call:
-            if self.reply_ctx.meta is not None:
-                parent_id = self.reply_ctx.meta.id  # slave meta id
-
         # tool output append
-        data = {
+        dispatch(RenderEvent(RenderEvent.TOOL_UPDATE, {
             "meta": self.reply_ctx.meta,
             "tool_data": tool_data,
-        }
-        event = RenderEvent(RenderEvent.TOOL_UPDATE, data)
-        self.window.dispatch(event)
+        }))
         self.clear()
 
         # disable reply if LlamaIndex agent is used
-        mode = self.window.core.config.get("mode")
-        if mode == MODE_LLAMA_INDEX:
-            if self.window.core.config.get("llama.idx.react", False):
-                self.window.core.debug.info("Reply disabled for LlamaIndex ReAct agent")
-                return
+        mode = core.config.get("mode")
+        if mode == MODE_LLAMA_INDEX and core.config.get("llama.idx.react", False):
+            return
 
         # send reply
         context = BridgeContext()
         context.ctx = prev_ctx
         context.prompt = str(tool_data)
-        extra = {
-            "force": True,
-            "reply": True,
-            "internal": True,
-            "parent_id": parent_id,
-        }
-        event = KernelEvent(KernelEvent.REPLY_RETURN, {
+        dispatch(KernelEvent(KernelEvent.REPLY_RETURN, {
             'context': context,
-            'extra': extra,
-        })
-        self.window.dispatch(event)
-
-    def run_post_response(
+            'extra': {
+                "force": True,
+                "reply": True,
+                "internal": True,
+            },
+        }))
+
+    def on_post_response(
             self,
             ctx: CtxItem,
             extra_data: Optional[Dict[str, Any]] = None
@@ -157,8 +149,7 @@ class Reply:
             if (ctx is None or not ctx.agent_call) or not self.window.controller.kernel.is_threaded():
                 if "post_update" in extra_data and isinstance(extra_data["post_update"], list):
                     if "file_explorer" in extra_data["post_update"]:
-                        # update file explorer view
-                        self.window.controller.files.update_explorer()
+                        self.window.controller.files.update_explorer()  # update file explorer view
 
     def clear(self):
         """Clear reply stack"""
@@ -172,8 +163,4 @@ class Reply:
 
         :return: true if can be logged
         """
-        is_log = False
-        if self.window.core.config.has("log.events") \
-                and self.window.core.config.get("log.events"):
-            is_log = True
-        return is_log
+        return self.window.core.config.get("log.events", False)

pygpt_net/controller/kernel/stack.py

@@ -6,12 +6,12 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.03 14:00:00                  #
+# Updated Date: 2025.08.23 15:00:00                  #
 # ================================================== #
 
 from typing import Any
 
-from qasync import QApplication
+from PySide6.QtWidgets import QApplication
 
 from pygpt_net.core.events import KernelEvent
 from pygpt_net.core.bridge.context import BridgeContext
@@ -44,14 +44,12 @@ class Stack:
         """
         Check if reply stack has any item
 
-        :return: True if has
+        :return: True if has items
         """
         return self.current is not None
 
     def clear(self):
-        """
-        Clear reply stack
-        """
+        """Clear reply stack"""
         self.current = None
         self.unlock()
 
@@ -71,35 +69,37 @@ class Stack:
                 context.parent_id,  # expert id
                 context.input,  # query
             )
+        # cmd execute
         elif context.type == ReplyContext.CMD_EXECUTE:
             self.window.controller.plugins.apply_cmds(
                 context.ctx,  # current ctx
                 context.cmds,  # commands
             )
+        # cmd execute (force)
         elif context.type == ReplyContext.CMD_EXECUTE_FORCE:
             self.window.controller.plugins.apply_cmds(
                 context.ctx,  # current ctx
                 context.cmds,  # commands
                 all=True,
             )
+        # cmd execute (inline)
         elif context.type == ReplyContext.CMD_EXECUTE_INLINE:
             self.window.controller.plugins.apply_cmds_inline(
                 context.ctx,  # current ctx
                 context.cmds,  # commands
             )
+        # agent continue
         elif context.type == ReplyContext.AGENT_CONTINUE:
             bridge_context = BridgeContext()
             bridge_context.ctx = context.ctx
-            bridge_context.prompt = context.input
-            extra = {
-                "force": True,
-                "internal": True,
-            }
-            event = KernelEvent(KernelEvent.INPUT_SYSTEM, {
+            bridge_context.prompt = context.input  # from reply context
+            self.window.dispatch(KernelEvent(KernelEvent.INPUT_SYSTEM, {
                 'context': bridge_context,
-                'extra': extra,
-            })
-            self.window.dispatch(event)
+                'extra': {
+                    "force": True,
+                    "internal": True,
+                },
+            }))
 
     def is_locked(self) -> bool:
         """
@@ -131,7 +131,7 @@ class Stack:
 
     def waiting(self) -> bool:
         """
-        Check if reply stack is waiting
+        Check if reply stack is waiting for processing
 
         :return: True if waiting
         """

pygpt_net/controller/model/importer.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.02 20:00:00                  #
+# Updated Date: 2025.08.24 20:00:00                  #
 # ================================================== #
 
 import copy
@@ -475,6 +475,7 @@ class Importer:
                         MODE_LLAMA_INDEX,
                         MODE_AGENT,
                         MODE_AGENT_LLAMA,
+                        MODE_AGENT_OPENAI,
                         MODE_EXPERT,
                     ]
                     m.provider = 'ollama'

pygpt_net/controller/model/model.py

@@ -6,18 +6,20 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.15 03:00:00                  #
+# Updated Date: 2025.08.23 15:00:00                  #
 # ================================================== #
 
+import os
 from typing import Optional
 
-from pygpt_net.core.events import Event, AppEvent
+from pygpt_net.core.events import Event, AppEvent, BaseEvent
+from pygpt_net.core.types import MODE_LLAMA_INDEX, MODE_CHAT
 from pygpt_net.item.model import ModelItem
+from pygpt_net.utils import trans
 
 from .editor import Editor
 from .importer import Importer
 
-
 class Model:
     def __init__(self, window=None):
         """
@@ -32,6 +34,63 @@ class Model:
     def _ensure_current_model_map(self):
         return self.window.core.config.data.setdefault('current_model', {})
 
+    def handle(self, event: BaseEvent):
+        """
+        Handle events
+
+        :param event: BaseEvent: Event to handle
+        """
+        name = event.name
+        mode = self.window.core.config.get("mode")
+
+        # on input begin
+        if name == Event.INPUT_BEGIN:
+            force = event.data.get("force", False)
+            stop = event.data.get("stop", False)
+            if not force and not stop:
+                # check ollama model
+                model = self.window.core.config.get('model')
+                if model:
+                    model_data = self.window.core.models.get(model)
+                    if model_data is not None and model_data.is_ollama():
+                        if (mode == MODE_LLAMA_INDEX or
+                                (
+                                        mode == MODE_CHAT and not model_data.is_openai_supported() and model_data.is_ollama()
+                                )
+                        ):
+                            model_id = model_data.get_ollama_model()
+
+                            # load ENV vars first
+                            if ('env' in model_data.llama_index
+                                    and model_data.llama_index['env'] is not None):
+                                for item in model_data.llama_index['env']:
+                                    key = item.get('name', '').strip()
+                                    value = item.get('value', '').strip()
+                                    os.environ[key] = value
+                            status = self.window.core.models.ollama.check_model(model_id)
+                            is_installed = status.get('is_installed', False)
+                            is_model = status.get('is_model', False)
+                            if not is_installed:
+                                event.data["stop"] = True  # stop flow
+                                self.window.ui.dialogs.alert(trans("dialog.ollama.not_installed"))
+                                return
+                            if not is_model:
+                                event.data["stop"] = True  # stop flow
+                                self.window.ui.dialogs.alert(
+                                    trans("dialog.ollama.model_not_found").replace("{model}", model_id))
+                                return
+
+        # on input before
+        elif name == Event.INPUT_BEFORE:
+            # check API key, show monit if no API key for current provider
+            model = self.window.core.config.get('model')
+            if model:
+                model_data = self.window.core.models.get(model)
+                if not self.window.controller.chat.common.check_api_key(mode=mode, model=model_data, monit=False):
+                    self.window.controller.chat.common.check_api_key(mode=mode, model=model_data, monit=True)
+                    event.data["stop"] = True  # stop flow
+                    return
+
     def select(self, model: str):
         """
         Select model

pygpt_net/controller/settings/editor.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.11 19:00:00                  #
+# Updated Date: 2025.08.24 02:00:00                  #
 # ================================================== #
 
 import copy
@@ -67,7 +67,7 @@ class Editor:
         self.window.ui.add_hook("update.config.notepad.num", self.hook_update)
         self.window.ui.add_hook("update.config.render.code_syntax", self.hook_update)
         self.window.ui.add_hook("update.config.theme.style", self.hook_update)
-        self.window.ui.add_hook("update.config.llama.idx.chat.agent.render.all", self.hook_update)
+        self.window.ui.add_hook("update.config.agent.output.render.all", self.hook_update)
         self.window.ui.add_hook("update.config.audio.input.backend", self.hook_update)
         self.window.ui.add_hook("update.config.audio.output.backend", self.hook_update)
         # self.window.ui.add_hook("llama.idx.storage", self.hook_update)  # vector store update
@@ -204,7 +204,7 @@ class Editor:
         if self.config_changed('ctx.sources') or self.config_changed('ctx.audio'):
             self.window.controller.ctx.refresh()
 
-        if self.config_changed('llama.idx.chat.agent.render.all'):
+        if self.config_changed('agent.output.render.all'):
             self.window.controller.chat.render.reload()
 
         # update global shortcuts
@@ -294,7 +294,7 @@ class Editor:
             self.window.core.config.set(key, value)
             self.window.controller.ctx.update()
 
-        elif key == "llama.idx.chat.agent.render.all":
+        elif key == "agent.output.render.all":
             self.window.core.config.set(key, value)
             self.window.controller.chat.render.reload()
 

pygpt_net/controller/ui/ui.py

@@ -6,20 +6,20 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.15 03:00:00                  #
+# Updated Date: 2025.08.23 15:00:00                  #
 # ================================================== #
 
 from typing import Optional
 
 from PySide6.QtGui import QColor
 
+from pygpt_net.core.events import BaseEvent, Event
 from pygpt_net.utils import trans
 
 from .mode import Mode
 from .tabs import Tabs
 from .vision import Vision
 
-
 class UI:
     def __init__(self, window=None):
         """
@@ -65,6 +65,20 @@ class UI:
         self.vision.update()
         self.window.controller.agent.legacy.update()
 
+    def handle(self, event: BaseEvent):
+        """
+        Handle events
+
+        :param event: BaseEvent: Event to handle
+        """
+        name = event.name
+
+        # on input begin
+        if name == Event.INPUT_BEGIN:
+            self.tabs.switch_to_first_chat()  # switch to first active chat tab
+        elif name == Event.CTX_END:
+            self.update_tokens()  # update UI
+
     def get_colors(self) -> dict:
         """
         Get color labels

pygpt_net/core/agents/observer/evaluation.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.12 00:00:00                  #
+# Updated Date: 2025.08.24 02:00:00                  #
 # ================================================== #
 
 from typing import List
@@ -188,8 +188,8 @@ class Evaluation:
         i = 0
         for ctx in history:
             # if next input (but not last) then clear outputs - use only output after last user input
-            if self.is_input(ctx) and i < len(history) - 1:
-                outputs.clear()
+            # if self.is_input(ctx) and i < len(history) - 1:
+                # outputs.clear()
 
             if self.is_output(ctx):
                 if ctx.output:

pygpt_net/core/agents/provider.py

@@ -6,11 +6,13 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.01 03:00:00                  #
+# Updated Date: 2025.08.24 03:00:00                  #
 # ================================================== #
+import os
+from typing import List, Dict, Any
 
-from typing import List, Dict
-
+from pygpt_net.core.types import MODE_CHAT
+from pygpt_net.item.model import ModelItem
 from pygpt_net.provider.agents.base import BaseAgent
 
 
@@ -94,3 +96,23 @@ class Provider:
         # sort by name
         choices.sort(key=lambda x: list(x.values())[0].lower())
         return choices
+
+    def get_openai_model(self, model: ModelItem) -> Any:
+        """
+        Get OpenAI model by model id
+
+        :param model: ModelItem
+        :return: OpenAI model provider
+        """
+        from openai import AsyncOpenAI
+        from agents import (
+            OpenAIChatCompletionsModel,
+        )
+        if model.provider == "openai":
+            return model.id
+        else:
+            args = self.window.core.models.prepare_client_args(MODE_CHAT, model)
+            return OpenAIChatCompletionsModel(
+                model=model.id,
+                openai_client=AsyncOpenAI(**args),
+            )

pygpt_net/core/agents/runner.py

@@ -105,6 +105,7 @@ class Runner:
             # vector store idx from preset
             if preset:
                 vector_store_idx = preset.idx
+                extra["agent_idx"] = vector_store_idx
 
             # tools
             agent_tools = self.window.core.agents.tools
@@ -126,7 +127,9 @@ class Runner:
 
             # --- ADDITIONAL CONTEXT ---
             # append additional context from RAG if available
-            if vector_store_idx and self.window.core.config.get("agent.idx.auto_retrieve", True):
+            if (vector_store_idx
+                    and vector_store_idx != "_"
+                    and self.window.core.config.get("agent.idx.auto_retrieve", True)):
                 ad_context = self.window.core.idx.chat.query_retrieval(
                     query=prompt,
                     idx=vector_store_idx,

pygpt_net/core/agents/runners/llama_workflow.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.14 01:00:00                  #
+# Updated Date: 2025.08.24 02:00:00                  #
 # ================================================== #
 
 import re
@@ -87,13 +87,14 @@ class LlamaWorkflow(BaseRunner):
             self.set_idle(signals)
             return False
 
+        use_current = False
         if ctx.partial:
+            use_current = True  # use current item as response if partial item (do not create new one)
             ctx.partial = False  # reset partial flag
 
-        response_ctx = self.make_response(ctx)
+        response_ctx = self.make_response(ctx, use_current=use_current)
         self.end_stream(response_ctx, signals)
         self.send_response(response_ctx, signals, KernelEvent.APPEND_DATA)  # send response
-
         self.set_idle(signals)
         return True
 
@@ -152,19 +153,24 @@ class LlamaWorkflow(BaseRunner):
             ctx.output = ctx.agent_final_response  # set output to current context
         else:
             ctx.output = ctx.live_output
-
         return ctx
 
     def make_response(
             self,
-            ctx: CtxItem
+            ctx: CtxItem,
+            use_current: bool = False
     ) -> CtxItem:
         """
         Create a response context item with the given input and output.
 
         :param ctx: CtxItem - the context item to use as a base
+        :param use_current: If True, use the current context item instead of creating a new one
         """
-        response_ctx = self.add_ctx(ctx, with_tool_outputs=True)
+        if use_current:
+            response_ctx = ctx  # use current context item
+        else:
+            response_ctx = self.add_ctx(ctx, with_tool_outputs=True)
+
         response_ctx.set_input("")
 
         prev_output = ctx.live_output
@@ -176,6 +182,9 @@ class LlamaWorkflow(BaseRunner):
         response_ctx.extra["agent_output"] = True  # mark as output response
         response_ctx.extra["agent_finish"] = True  # mark as finished
 
+        if "agent_input" in response_ctx.extra:
+            del response_ctx.extra["agent_input"]  # remove agent input from extra
+
         if ctx.agent_final_response:  # only if not empty
             response_ctx.extra["output"] = ctx.agent_final_response
 
@@ -194,9 +203,12 @@ class LlamaWorkflow(BaseRunner):
         :param output: Output string
         :return: Filtered output string
         """
+        if output is None:
+            return ""
+
         # Remove <execute>...</execute> tags
         filtered_output = re.sub(r'<execute>.*?</execute>', '', output, flags=re.DOTALL)
-        return filtered_output
+        return filtered_output.strip()
 
     async def run_agent(
             self,

pygpt_net/core/agents/runners/loop.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2025.08.14 03:00:00                  #
+# Updated Date: 2025.08.24 02:00:00                  #
 # ================================================== #
 
 from typing import Optional, List
@@ -136,6 +136,7 @@ class Loop(BaseRunner):
         self.set_status(signals, msg)
         if score < 0:
             self.send_response(ctx, signals, KernelEvent.APPEND_END)
+            self.set_idle(signals)
             return True
         good_score = self.window.core.config.get("agent.llama.loop.score", 75)
         if score >= good_score != 0:
@@ -145,6 +146,7 @@ class Loop(BaseRunner):
                 score=str(score)
             )
             self.send_response(ctx, signals, KernelEvent.APPEND_END, msg=msg)
+            self.set_idle(signals)
             return True
 
         # print("Instruction: " + instruction, "Score: " + str(score))