Deleted memory tool

app.py

@@ -2,8 +2,6 @@ import os, threading
 import gradio as gr
 from crew import run_crew
 from utils import get_questions
-from tools.memory_tools import memory_tool
-
 
 
 def ask(question, openai_api_key, gemini_api_key, anthropic_api_key, file_name = ""):

crew.py

@@ -4,7 +4,6 @@
 # https://ai.google.dev/gemini-api/docs
 
 import os
-from tools.memory_tools import memory_tool
 from crewai import Agent, Crew, Task
 from crewai.agents.agent_builder.base_agent import BaseAgent
 from crewai.project import CrewBase, agent, crew, task
@@ -122,8 +121,7 @@ class GAIACrew():
                 verbose=True
             ))
         return agents
-    
-    
+        
     @task
     def manager_task(self) -> Task:
         # Build the Task object from your YAML
@@ -139,8 +137,6 @@ class GAIACrew():
         
         return task
 
-
-
     def get_crew(self) -> Crew:
         return Crew(
             agents=self.agents,
@@ -149,7 +145,6 @@ class GAIACrew():
         )
 
 
-
 def run_crew(question, file_path):
     # 0) Prepend file data if needed
     final_question = question

faiss_index/index.pkl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:04bd4a34c4fb5c05840d6d26afa952c49bad6fd9e93951096a45584c4ad69dbb
-size 893

tools/memory_tools.py

@@ -1,91 +0,0 @@
-# tools/memory_tools.py
-
-import os
-from crewai.tools import tool
-from langchain.schema import Document
-from langchain.embeddings import OpenAIEmbeddings
-from langchain_community.vectorstores import FAISS   # faiss-cpu must be installed
-
-INDEX_DIR = "faiss_index"
-_embeddings = None   # will hold our OpenAIEmbeddings
-_vectorstore = None  # will hold our FAISS instance
-
-def _init_embeddings():
-    global _embeddings
-    if _embeddings is None:
-        _embeddings = OpenAIEmbeddings(
-            openai_api_key=os.environ.get("OPENAI_API_KEY", "")
-        )
-    return _embeddings
-
-def _load_or_build_index(doc: Document = None):
-    """
-    - If there's an in-memory index but the on-disk folder was removed, reset it.
-    - If there's an in-memory index and the folder still exists, reuse it.
-    - Else, if there's an on-disk index, load it.
-    - Else, if a single `doc` is provided, create a new index from it.
-    """
-    global _vectorstore
-    emb = _init_embeddings()
-
-    # 1) If we had an in-memory index but the INDEX_DIR was deleted, clear it so we rebuild
-    if _vectorstore is not None and not os.path.isdir(INDEX_DIR):
-        _vectorstore = None
-
-    # 2) If we now have an in-memory index, reuse it
-    if _vectorstore is not None:
-        return _vectorstore
-
-    # 3) On‐disk index?
-    if os.path.isdir(INDEX_DIR):
-        _vectorstore = FAISS.load_local(
-            INDEX_DIR, emb, allow_dangerous_deserialization=True
-        )
-        return _vectorstore
-
-    # 4) No index yet, but we're saving the first doc
-    if doc is not None:
-        _vectorstore = FAISS.from_documents([doc], emb)
-        _vectorstore.save_local(INDEX_DIR)
-        return _vectorstore
-
-    # 5) Otherwise, no index and no doc to build from
-    return None
-
-@tool("Memory Tool")
-def memory_tool(action: str, text: str) -> str:
-    """
-    action: "save" to store the user message, or "load" to retrieve similar past messages.
-    text:   the message to save, or the query for load.
-
-    Returns:
-      - on "save": "Saved"
-      - on "load": up to 3 similar messages joined by newline, or "" if none
-      - otherwise: "Invalid action"
-    """
-    act = action.strip().lower()
-
-    if act == "save":
-        # Wrap the text in a Document
-        doc = Document(page_content=text)
-
-        # Build or load the index (if it's the first doc, we pass it here)
-        vs = _load_or_build_index(doc)
-
-        # If we already had an index, just add the new doc
-        if vs and os.path.isdir(INDEX_DIR):
-            vs.add_documents([doc])
-            vs.save_local(INDEX_DIR)
-
-        return "Saved"
-
-    elif act == "load":
-        vs = _load_or_build_index()
-        if not vs:
-            return ""    # no history yet
-
-        hits = vs.similarity_search(text, k=3)
-        return "\n".join(d.page_content for d in hits)
-
-    else:
-        return "Invalid action"