Updated dependencies in pyproject.toml and uv.lock, including upgradi…

pyproject.toml

@@ -6,7 +6,7 @@ authors = [{ name = "Your Name", email = "you@example.com" }]
 requires-python = ">=3.10,<3.14"
 dependencies = [
     "colorama>=0.4.6",
-    "crewai[tools]>=0.134.0,<1.0.0",
+    "crewai[tools]==0.201.1",
     "streamlit>=1.44.1",
 ]
 

src/conversational_routing/crews/assistant_crew/assistant_crew.py

@@ -1,10 +1,6 @@
-import os
-import glob
 from crewai import Agent, Crew, Process, Task, LLM
 from crewai.project import CrewBase, agent, crew, task
-from crewai.knowledge.source.text_file_knowledge_source import TextFileKnowledgeSource
-from crewai.knowledge.knowledge_config import KnowledgeConfig
-from pathlib import Path
+
 
 @CrewBase
 class AssistantCrew:
@@ -16,30 +12,14 @@ class AssistantCrew:
 
     # LLM Configuration
     llm = LLM(
-        model="groq/meta-llama/llama-4-scout-17b-16e-instruct",
+        model="gpt-4o-mini",
         temperature=0.1,
     )
 
     @agent
     def crewai_expert_agent(self) -> Agent:
-        # Knowledge Configuration
-        # Define base path to current file
-        knowledge_base_path = Path(__file__).parent / "knowledge"
-
-        # Prepare the knowledge base for the OSS Framework
-        files = glob.glob(os.path.join(knowledge_base_path, "oss-docs/**/*.mdx"), recursive=True)
-        # Convert file strings to Path objects
-        knowledge_file_paths = [Path(file) for file in files]
-
         return Agent(
             config=self.agents_config["crewai_expert_agent"],
-            knowledge_sources=[TextFileKnowledgeSource(
-                file_paths=knowledge_file_paths,
-                metadata={
-                    "category": "CrewAI",
-                },
-            )],
-            knowledge_config=KnowledgeConfig(results_limit=5, score_threshold=0.7),
             llm=self.llm,
         )
 
@@ -57,5 +37,5 @@ def crew(self) -> Crew:
             tasks=self.tasks,  # Automatically created by the @task decorator
             process=Process.sequential,
             verbose=True,
-            memory=True,
+            # memory=True,  # Enables short-term, long-term, and entity memory
         )

src/conversational_routing/crews/assistant_crew/config/agents.yaml

@@ -1,16 +1,11 @@
 crewai_expert_agent:
   role: >
-    CrewAI Support Expert
+    Helpful AI Assistant
   goal: >
-    Provide concise (less than a 500 characters unless there is a code example) and accurate information about the CrewAI framework, answering user questions and guiding them in its effective use.
-    Provide code snippets if required.
+    Have natural conversations with users. Be friendly and helpful.
+    If users ask about CrewAI, provide accurate information and code examples.
+    If users ask about other topics, respond naturally and conversationally.
   backstory: >
-    You are a helpful and knowledgeable CrewAI expert. 
-    You have extensive experience with the framework and its various components. 
-    You are dedicated to assisting users in understanding and utilizing CrewAI for their agentic AI projects.
-    You are familiar with the latest features, best practices, and community resources related to CrewAI.
-    You are passionate about helping others build successful agentic AI solutions using CrewAI.
-    You stay up-to-date with the latest developments in the CrewAI ecosystem.
-    You are adept at explaining complex concepts in a clear and concise manner.
-    You are patient and willing to go the extra mile to ensure users grasp the intricacies of CrewAI. 
-    You are always eager to share your expertise and contribute to the growth of the CrewAI community.
+    You are a friendly AI assistant who can chat about various topics.
+    You're particularly knowledgeable about CrewAI but happy to have casual conversations too.
+    You remember what users tell you during the conversation.

src/conversational_routing/crews/assistant_crew/config/tasks.yaml

@@ -1,11 +1,11 @@
 answer_crewai_questions_task:
   agent: crewai_expert_agent
   description: >
-    Respond to user questions about the CrewAI framework.
+    Have a natural conversation with the user. Respond appropriately to their message.
+    Use the context provided to maintain conversation continuity.
 
+    Previous context summary: {conversation_summary}
+    Recent messages: {conversation_history}
     Current message: {current_message}
-    Conversation history: {conversation_history}
   expected_output: >
-    If required:
-    - Provide links to relevant documentation
-    - Offer practical examples
+    A helpful, conversational response that addresses what the user said.

src/conversational_routing/main.py

@@ -4,21 +4,79 @@
 import json
 
 from crewai.flow import Flow, start, persist
+from litellm import completion
 
 from src.conversational_routing.crews.assistant_crew.assistant_crew import AssistantCrew
 
+
 class ChatState(BaseModel):
     current_message: str = ""
     conversation_history: List[dict] = []
+    conversation_summary: str = ""  # Running summary of older context
+    summary_threshold: int = 8      # Summarize when history exceeds this
+    keep_recent: int = 4            # Keep this many recent messages in full
 
 @persist()
 class ChatFlow(Flow[ChatState]):
+
+    def _summarize_conversation(self) -> None:
+        """
+        Summarize older messages when conversation history exceeds threshold.
+        Keeps recent messages in full detail while compressing older context.
+        """
+        history_length = len(self.state.conversation_history)
+
+        if history_length <= self.state.summary_threshold:
+            return  # No need to summarize yet
+
+        # Calculate how many messages to summarize
+        messages_to_summarize = history_length - self.state.keep_recent
+        if messages_to_summarize <= 0:
+            return
+
+        # Extract older messages to summarize
+        older_messages = self.state.conversation_history[:messages_to_summarize]
+
+        # Format messages for summarization
+        formatted_messages = "\n".join([
+            f"{msg['role'].upper()}: {msg['content']}" 
+            for msg in older_messages
+        ])
+
+        # Include existing summary if present
+        existing_context = ""
+        if self.state.conversation_summary:
+            existing_context = f"Previous summary: {self.state.conversation_summary}\n\n"
+
+        # Use lightweight LLM call for summarization
+        summary_prompt = f"""Summarize the following conversation context concisely, preserving key facts, user preferences, and important details mentioned. Keep it under 200 words.
+
+{existing_context}New messages to incorporate:
+{formatted_messages}
+
+Concise summary:"""
+
+        response = completion(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": summary_prompt}],
+            temperature=0.3,
+            max_tokens=300
+        )
+
+        # Update summary and trim conversation history
+        self.state.conversation_summary = response.choices[0].message.content
+        self.state.conversation_history = self.state.conversation_history[messages_to_summarize:]
+
     @start()
     def answer_message(self):
+        # Check if we need to summarize before processing
+        self._summarize_conversation()
+
         # Here define the crew that will respond to the user message
         assistant_crew = AssistantCrew().crew().kickoff({
             "current_message": self.state.current_message,
-            "conversation_history": self.state.conversation_history
+            "conversation_history": self.state.conversation_history,
+            "conversation_summary": self.state.conversation_summary
         })
 
         response = assistant_crew.raw