Fixed auto-ctx indexing in autonomous mode - issue #23

src/pygpt_net/controller/agent/flow.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2024.03.07 23:00:00                  #
+# Updated Date: 2024.03.08 23:00:00                  #
 # ================================================== #
 
 from pygpt_net.item.ctx import CtxItem
@@ -46,12 +46,17 @@ def on_system_prompt(
         """
         stop_cmd = ""
         if auto_stop:
-            stop_cmd = '\n\nSTATUS UPDATE: You can use "goal_update" command to update status of the task. Remember to put it in the form as given, at the end of response and including ' \
-                       'the surrounding ~###~ marks, e.g.: ~###~{"cmd": "goal_update", "params": {"status": "finished"}}~###~'
+            stop_cmd = '\n\nSTATUS UPDATE: You can use "goal_update" command to update status of the task. ' \
+                       'Remember to put it in the form as given, at the end of response and including ' \
+                       'the surrounding ~###~ marks, ' \
+                       'e.g.: ~###~{"cmd": "goal_update", "params": {"status": "finished"}}~###~'
             stop_cmd+= '\nON GOAL FINISH: When you believe that the task has been completed 100% and all goals have ' \
                        'been achieved, run "goal_update" command with status = "finished".'
-            stop_cmd += '\nON PAUSE, FAILED OR WAIT: If more data from user is needed to achieve the goal or task run must be paused or ' \
-                        'task was failed or when the conversation falls into a loop, THEN STOP REASONING and include "goal_update" command with one of these statuses: "pause", "failed" or "wait"'
+            stop_cmd += '\nON PAUSE, FAILED OR WAIT: If more data from user is needed to achieve the goal or task ' \
+                        'run must be paused or ' \
+                        'task was failed or when the conversation falls into a loop, ' \
+                        'THEN STOP REASONING and include "goal_update" command with one of these statuses: ' \
+                        '"pause", "failed" or "wait"'
         if append_prompt is not None and append_prompt.strip() != "":
             append_prompt = "\n" + append_prompt
         prompt += str(append_prompt) + stop_cmd
@@ -109,9 +114,9 @@ def on_ctx_end(
                     force=True,
                     internal=True,
                 )
-            # internal call will not trigger async mode and will hide the message from previous iteration
+        # internal call will not trigger async mode and will hide the message from previous iteration
         elif self.iteration >= int(iterations):
-            self.on_stop()
+            self.on_stop(auto=True)
             if self.window.core.config.get("agent.goal.notify"):
                 self.window.ui.tray.show_msg(
                     trans("notify.agent.stop.title"),
@@ -194,7 +199,7 @@ def cmd(
             try:
                 if item["cmd"] == "goal_update":
                     if item["params"]["status"] == "finished":
-                        self.on_stop()
+                        self.on_stop(auto=True)
                         self.window.ui.status(trans('status.finished'))  # show info
                         self.finished = True
                         if self.window.core.config.get("agent.goal.notify"):
@@ -203,19 +208,25 @@ def cmd(
                                 trans("notify.agent.goal.content"),
                             )
                     elif item["params"]["status"] in pause_status:
-                        self.on_stop()
+                        self.on_stop(auto=True)
                         self.window.ui.status(trans('status.finished'))  # show info
                         self.finished = True
             except Exception as e:
                 self.window.core.debug.error(e)
                 return
 
-    def on_stop(self):
+    def on_stop(self, auto: bool = False):
         """
         Event: On force stop
+
+        :param auto: auto
         """
         self.window.controller.chat.common.unlock_input()
         self.iteration = 0
         self.prev_output = None
         self.stop = True
         self.finished = False  # reset finished flag
+
+        # update index if auto-index enabled, sync mode
+        if auto:
+            self.window.controller.idx.on_ctx_end(None, sync=True)

src/pygpt_net/controller/chat/input.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2024.02.25 06:00:00                  #
+# Updated Date: 2024.03.08 23:00:00                  #
 # ================================================== #
 
 from PySide6.QtWidgets import QApplication
@@ -33,7 +33,8 @@ def __init__(self, window=None):
         self.no_ctx_idx_modes = [
             "img",
             "assistant",
-            "llama_index"
+            "llama_index",
+            "agent",
         ]  # assistant handled in async
 
     def send_input(self, force: bool = False):
@@ -230,8 +231,9 @@ def execute(
         self.window.controller.ui.update_tokens()  # update UI
         self.generating = False  # unlock
 
-        if mode not in self.no_ctx_idx_modes:
+        if mode not in self.no_ctx_idx_modes and not self.window.controller.agent.enabled():
             self.window.controller.idx.on_ctx_end(ctx)  # update ctx DB index
+            # disabled in agent mode to prevent load everything at once when runs finished.
 
         self.log("End.")
 

src/pygpt_net/controller/ctx/common.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2024.03.06 02:00:00                  #
+# Updated Date: 2024.03.08 23:00:00                  #
 # ================================================== #
 
 from PySide6.QtWidgets import QApplication
@@ -132,7 +132,6 @@ def restore_filters_labels(self):
             self.window.core.ctx.filters_labels = labels
             self.window.ui.nodes['filter.ctx.labels'].restore(labels)
 
-
     def toggle_display_filter(self, filter: str):
         """
         Toggle display filter

src/pygpt_net/controller/idx/__init__.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2024.02.23 01:00:00                  #
+# Updated Date: 2024.03.08 23:00:00                  #
 # ================================================== #
 
 import datetime
@@ -141,11 +141,12 @@ def update_list(self):
         if items is not None:
             self.window.ui.toolbox.indexes.update(items)
 
-    def on_ctx_end(self, ctx: CtxItem = None):
+    def on_ctx_end(self, ctx: CtxItem = None, sync: bool = False):
         """
         After context item updated (request + response received)
 
         :param ctx: Context item instance
+        :param sync: Synchronous call
         """
         # ignore if manually stopped
         if self.window.controller.chat.input.stop:
@@ -161,7 +162,7 @@ def on_ctx_end(self, ctx: CtxItem = None):
             if current_ctx is not None:
                 meta = self.window.core.ctx.get_meta_by_id(current_ctx)
                 if meta is not None:
-                    self.indexer.index_ctx_realtime(meta, idx)
+                    self.indexer.index_ctx_realtime(meta, idx, sync=sync)
 
     def after_index(self, idx: str = None):
         """

src/pygpt_net/core/idx/chat.py

@@ -6,7 +6,7 @@
 # GitHub:  https://github.com/szczyglis-dev/py-gpt   #
 # MIT License                                        #
 # Created By  : Marcin Szczygliński                  #
-# Updated Date: 2024.03.04 20:00:00                  #
+# Updated Date: 2024.03.08 23:00:00                  #
 # ================================================== #
 
 import json
@@ -182,7 +182,7 @@ def chat(self, **kwargs) -> bool:
                 ctx.stream = response.response_gen
                 ctx.input_tokens = input_tokens
                 ctx.set_output("", "")
-                # no metadata here in agent chat stream response
+                # no metadata here in agent chat stream response!
             else:
                 ctx.add_doc_meta(response.metadata)  # store metadata
                 ctx.input_tokens = input_tokens
@@ -196,48 +196,65 @@ def chat(self, **kwargs) -> bool:
             return True
         return False
 
-    def query_file(self, path: str, query: str) -> str:
+    def query_file(self, path: str, query: str, model: ModelItem = None) -> str:
         """
         Query file using temp index (created on the fly)
 
-        :param path: path to file to index
+        :param path: path to file to index (in memory)
         :param query: query
+        :param model: model
+        :return: response
         """
-        model = self.window.core.models.from_defaults()
+        if model is None:
+            model = self.window.core.models.from_defaults()
         context = self.window.core.idx.llm.get_service_context(model=model)
-        index = self.storage.get_tmp(path, service_context=context)  # get or create tmp index
+        tmp_id, index = self.storage.get_tmp(path, service_context=context)  # get or create tmp index
 
         idx = "tmp:{}".format(path)  # tmp index id
         self.log("Indexing to temporary in-memory index: {}...".format(idx))
 
         # index file to tmp index
-        files, errors = self.window.core.idx.indexing.index_files(idx, index, path, is_tmp=True)
+        files, errors = self.window.core.idx.indexing.index_files(
+            idx=idx,
+            index=index,
+            path=path,
+            is_tmp=True,  # do not try to remove old doc id
+        )
+
+        # query tmp index
+        output = None
         if len(files) > 0:
             self.log("Querying temporary in-memory index: {}...".format(idx))
             response = index.as_query_engine(
                 streaming=False,
             ).query(query)  # query with default prompt
             if response:
-                return response.response
+                output = response.response
 
-    def query_web(self, type: str, url: str, args: dict, query: str) -> str:
+        self.storage.clean_tmp(tmp_id)  # clean memory
+        return output
+
+    def query_web(self, type: str, url: str, args: dict, query: str, model: ModelItem = None) -> str:
         """
         Query web using temp index (created on the fly)
 
         :param type: type of content
-        :param url: url to index
+        :param url: url to index (in memory)
         :param args: extra args
         :param query: query
+        :param model: model
+        :return: response
         """
         parts = {
             "type": type,
             "url": url,
             "args": args,
         }
         id = json.dumps(parts)
-        model = self.window.core.models.from_defaults()
+        if model is None:
+            model = self.window.core.models.from_defaults()
         context = self.window.core.idx.llm.get_service_context(model=model)
-        index = self.storage.get_tmp(id, service_context=context)  # get or create tmp index
+        tmp_id, index = self.storage.get_tmp(id, service_context=context)  # get or create tmp index
 
         idx = "tmp:{}".format(id)  # tmp index id
         self.log("Indexing to temporary in-memory index: {}...".format(idx))
@@ -249,15 +266,21 @@ def query_web(self, type: str, url: str, args: dict, query: str) -> str:
             urls=[url],
             type=type,
             extra_args=args,
-            is_tmp = True,
+            is_tmp=True,  # do not try to remove old doc id
         )
+
+        # query tmp index
+        output = None
         if num > 0:
             self.log("Querying temporary in-memory index: {}...".format(idx))
             response = index.as_query_engine(
                 streaming=False,
             ).query(query)  # query with default prompt
             if response:
-                return response.response
+                output = response.response
+
+        self.storage.clean_tmp(tmp_id)  # clean memory
+        return output
 
     def get_memory_buffer(self, history: list, llm = None) -> ChatMemoryBuffer:
         """