diff --git a/.gitignore b/.gitignore
index 3e57a39..ef79b3a 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,3 +1,4 @@
 test-results/
 **.log
-**/report.html
\ No newline at end of file
+**/report.html
+docker-compose
\ No newline at end of file
diff --git a/app-backend/templates/tools/supervisor_agent_tools.yaml b/app-backend/templates/tools/supervisor_agent_tools.yaml
deleted file mode 100644
index 30d76f0..0000000
--- a/app-backend/templates/tools/supervisor_agent_tools.yaml
+++ /dev/null
@@ -1,20 +0,0 @@
- # Copyright (C) 2024 Intel Corporation
- # SPDX-License-Identifier: Apache-2.0
- 
- search_knowledge_base:
-   description: Search a knowledge base for a given query. Returns text related to the query.
-   callable_api: tools.py:search_knowledge_base
-   args_schema:
-     query:
-       type: str
-       description: query
-   return_output: retrieved_data
- 
- search_sql_database:
-   description: Search a SQL database with a natural language query. Returns text related to the query.
-   callable_api: tools.py:search_sql_database
-   args_schema:
-     query:
-       type: str
-       description: natural language query
-   return_output: retrieved_data
\ No newline at end of file
diff --git a/app-backend/templates/tools/tools.py b/app-backend/templates/tools/tools.py
deleted file mode 100644
index 6a0b643..0000000
--- a/app-backend/templates/tools/tools.py
+++ /dev/null
@@ -1,35 +0,0 @@
-# Copyright (C) 2024 Intel Corporation
- # SPDX-License-Identifier: Apache-2.0
- 
-import os
-import requests
-
-
-def search_knowledge_base(query: str) -> str:
-    """Search a knowledge base about music and singers for a given query.
-
-    Returns text related to the query.
-    """
-    url = os.environ.get("WORKER_AGENT_URL")
-    print(url)
-    proxies = {"http": ""}
-    payload = {
-        "messages": query,
-    }
-    response = requests.post(url, json=payload, proxies=proxies)
-    return response.json()["text"]
-
-
-def search_sql_database(query: str) -> str:
-    """Search a SQL database on artists and their music with a natural language query.
-
-    Returns text related to the query.
-    """
-    url = os.environ.get("SQL_AGENT_URL")
-    print(url)
-    proxies = {"http": ""}
-    payload = {
-        "messages": query,
-    }
-    response = requests.post(url, json=payload, proxies=proxies)
-    return response.json()["text"]
\ No newline at end of file
diff --git a/app-backend/templates/tools/worker_agent_tools.py b/app-backend/templates/tools/worker_agent_tools.py
deleted file mode 100644
index 43c5647..0000000
--- a/app-backend/templates/tools/worker_agent_tools.py
+++ /dev/null
@@ -1,40 +0,0 @@
-# Copyright (C) 2024 Intel Corporation
- # SPDX-License-Identifier: Apache-2.0
- 
-import os
-import requests
-
-def search_knowledge_base(query: str) -> str:
-    """Search the knowledge base for a specific query."""
-    url = os.environ.get("RETRIEVAL_TOOL_URL")
-    print(url)
-    proxies = {"http": ""}
-    payload = {
-        "text": query,
-    }
-    response = requests.post(url, json=payload, proxies=proxies)
-    print(response)
-    if "documents" in response.json():
-        docs = response.json()["documents"]
-        context = ""
-        for i, doc in enumerate(docs):
-            if i == 0:
-                context = doc
-            else:
-                context += "\n" + doc
-        # print(context)
-        return context
-    elif "text" in response.json():
-        return response.json()["text"]
-    elif "reranked_docs" in response.json():
-        docs = response.json()["reranked_docs"]
-        context = ""
-        for i, doc in enumerate(docs):
-            if i == 0:
-                context = doc["text"]
-            else:
-                context += "\n" + doc["text"]
-        # print(context)
-        return context
-    else:
-        return "Error parsing response from the knowledge base."
\ No newline at end of file
diff --git a/app-backend/templates/tools/worker_agent_tools.yaml b/app-backend/templates/tools/worker_agent_tools.yaml
deleted file mode 100644
index 1a0975e..0000000
--- a/app-backend/templates/tools/worker_agent_tools.yaml
+++ /dev/null
@@ -1,11 +0,0 @@
-# Copyright (C) 2024 Intel Corporation
-# SPDX-License-Identifier: Apache-2.0
-
-search_knowledge_base:
-  description: Search knowledge base for a given query. Returns text related to the query.
-  callable_api: worker_agent_tools.py:search_knowledge_base
-  args_schema:
-    query:
-      type: str
-      description: query
-  return_output: retrieved_data
\ No newline at end of file
diff --git a/app-frontend/react/src/components/Conversation/Conversation.tsx b/app-frontend/react/src/components/Conversation/Conversation.tsx
index 6ee4499..716cb30 100644
--- a/app-frontend/react/src/components/Conversation/Conversation.tsx
+++ b/app-frontend/react/src/components/Conversation/Conversation.tsx
@@ -41,6 +41,7 @@ const Conversation = ({ title, enabledUiFeatures }: ConversationProps) => {
   const [startTime, setStartTime] = useState<number | null>(null);
   const [isAssistantTyping, setIsAssistantTyping] = useState<boolean>(false);
   const [showInferenceParams, setShowInferenceParams] = useState<boolean>(true);
+  // const [isInThinkMode, setIsInThinkMode] = useState<boolean>(false);
 
   const toSend = "Enter";
 
@@ -74,6 +75,7 @@ const Conversation = ({ title, enabledUiFeatures }: ConversationProps) => {
       maxTokens: tokenLimit,
       temperature: temperature,
       model: "Intel/neural-chat-7b-v3-3",
+      // setIsInThinkMode
     });
     setPrompt("");
     setStartTime(Date.now());
@@ -89,24 +91,38 @@ const Conversation = ({ title, enabledUiFeatures }: ConversationProps) => {
       let tokenLength: number;
       if (isAgent) {
         const currentSteps = getCurrentAgentSteps();
-        const allContent = currentSteps.flatMap(step => step.content).join(" ");
-        tokenLength = allContent.split(" ").length;
+        const stepsContent = currentSteps.flatMap(step => step.content).join(" ");
+        const stepsSource = currentSteps.flatMap(step => step.source).join(" ");
+        const allContent = [stepsContent, stepsSource, onGoingResult].filter(str => str.trim()).join(" ");
+        let prevTokenLen = messageTokenData[`${selectedConversationId}-${currentMessageIndex}`]?.tokens || 0;
+        tokenLength = allContent.split(/\s+/).filter(token => token.length > 0).length + prevTokenLen;
+  
+        console.log("Agent Token Calc:", {
+          stepsContent,
+          stepsSource,
+          onGoingResult,
+          tokenLength
+        });
       } else {
-        tokenLength = onGoingResult.split(" ").length;
+        tokenLength = onGoingResult.split(/\s+/).filter(token => token.length > 0).length;
       }
-
+  
       const currentTimestamp = Date.now();
       const elapsedTime = (currentTimestamp - startTime) / 1000;
       const tokenRate = elapsedTime > 0 ? tokenLength / elapsedTime : 0;
-
-      setMessageTokenData((prev) => ({
-        ...prev,
-        [`${selectedConversationId}-${currentMessageIndex}`]: { tokens: tokenLength, rate: tokenRate, time: elapsedTime },
-      }));
-
+  
+      setMessageTokenData((prev) => {
+        const updatedData = {
+          ...prev,
+          [`${selectedConversationId}-${currentMessageIndex}`]: { tokens: tokenLength, rate: tokenRate, time: elapsedTime },
+        };
+        console.log("Updated token data:", updatedData);
+        return updatedData;
+      });
+  
       setIsAssistantTyping(false);
     }
-
+  
     scrollToBottom();
   }, [onGoingResult, startTime, selectedConversation?.Messages, currentMessageIndex, isAgent]);
 
@@ -180,6 +196,7 @@ const Conversation = ({ title, enabledUiFeatures }: ConversationProps) => {
                   tokenCount={message.role === MessageRole.Assistant ? tokens : undefined}
                   tokenRate={message.role === MessageRole.Assistant ? rate : undefined}
                   agentSteps={message.agentSteps || []}
+                  // isInThink={isInThinkMode}
                 />
               );
             })}
@@ -194,6 +211,7 @@ const Conversation = ({ title, enabledUiFeatures }: ConversationProps) => {
                 tokenCount={0}
                 tokenRate={0}
                 agentSteps={getCurrentAgentSteps()}
+                // isInThink={isInThinkMode}
               />
             )}
 
@@ -207,40 +225,40 @@ const Conversation = ({ title, enabledUiFeatures }: ConversationProps) => {
                 tokenCount={messageTokenData[`${selectedConversationId}-${currentMessageIndex}`]?.tokens}
                 tokenRate={messageTokenData[`${selectedConversationId}-${currentMessageIndex}`]?.rate}
                 agentSteps={getCurrentAgentSteps()}
+                // isInThink={isInThinkMode}
               />
             )}
           </div>
 
           <div className={styleClasses.conversatioSliders}>
-          <Button
-            variant="light"
-            size="xs"
-            radius="xl"
-            onClick={() => setShowInferenceParams(!showInferenceParams)}
-            rightSection={showInferenceParams ? <IconChevronDown size={14} /> : <IconChevronUp size={14} />}
-            mb="xs"
-          >
-            {showInferenceParams ? "Hide Inference Settings" : "Show Inference Settings"}
-          </Button>
-          <Collapse in={showInferenceParams} mb="md">
-            <Stack style={{ marginLeft: '10px' }}>
-              <Title size="sm">Inference Settings</Title>
-              <Text size="sm">Token Limit: {tokenLimit}</Text>
-              <Slider value={tokenLimit} onChange={setTokenLimit} min={10} max={500} step={1} />
-              <Text size="sm">Temperature: {temperature.toFixed(2)}</Text>
-              <Slider value={temperature} onChange={setTemperature} min={0.10} max={1.00} step={0.01} />
-              <Textarea
-                label="System Prompt"
-                placeholder="Set system prompt"
-                value={systemPrompt}
-                onChange={(e) => setSystemPrompt(e.target.value)}
-                size="sm"
-                mb="sm"
-              />
-            </Stack>
-          </Collapse>
-        </div>
-
+            <Button
+              variant="light"
+              size="xs"
+              radius="xl"
+              onClick={() => setShowInferenceParams(!showInferenceParams)}
+              rightSection={showInferenceParams ? <IconChevronDown size={14} /> : <IconChevronUp size={14} />}
+              mb="xs"
+            >
+              {showInferenceParams ? "Hide Inference Settings" : "Show Inference Settings"}
+            </Button>
+            <Collapse in={showInferenceParams} mb="md">
+              <Stack style={{ marginLeft: '10px' }}>
+                <Title size="sm">Inference Settings</Title>
+                <Text size="sm">Token Limit: {tokenLimit}</Text>
+                <Slider value={tokenLimit} onChange={setTokenLimit} min={10} max={500} step={1} />
+                <Text size="sm">Temperature: {temperature.toFixed(2)}</Text>
+                <Slider value={temperature} onChange={setTemperature} min={0.10} max={1.00} step={0.01} />
+                <Textarea
+                  label="System Prompt"
+                  placeholder="Set system prompt"
+                  value={systemPrompt}
+                  onChange={(e) => setSystemPrompt(e.target.value)}
+                  size="sm"
+                  mb="sm"
+                />
+              </Stack>
+            </Collapse>
+          </div>
 
           <div className={styleClasses.conversationActions}>
             <Tooltip
@@ -261,7 +279,7 @@ const Conversation = ({ title, enabledUiFeatures }: ConversationProps) => {
                     <IconArrowRight style={{ width: rem(18), height: rem(18) }} stroke={1.5} />
                   </ActionIcon>
                 }
-                disabled={!enabledUiFeatures.chat || !!onGoingResult}
+                disabled={!enabledUiFeatures.chat || !!onGoingResult || isAssistantTyping}
               />
             </Tooltip>
           </div>
diff --git a/app-frontend/react/src/components/Message/conversationMessage.tsx b/app-frontend/react/src/components/Message/conversationMessage.tsx
index bd53ffe..66df29d 100644
--- a/app-frontend/react/src/components/Message/conversationMessage.tsx
+++ b/app-frontend/react/src/components/Message/conversationMessage.tsx
@@ -19,6 +19,7 @@ export interface ConversationMessageProps {
   tokenRate?: number;
   elapsedTime?: number;
   agentSteps: AgentStep[];
+  // isInThink: boolean;
 }
 
 export function ConversationMessage({ human, message, date, elapsedTime, tokenCount, tokenRate, agentSteps }: ConversationMessageProps) {
diff --git a/app-frontend/react/src/redux/Conversation/Conversation.ts b/app-frontend/react/src/redux/Conversation/Conversation.ts
index 835a049..96ef58e 100644
--- a/app-frontend/react/src/redux/Conversation/Conversation.ts
+++ b/app-frontend/react/src/redux/Conversation/Conversation.ts
@@ -8,6 +8,7 @@ export type ConversationRequest = {
   model: string;
   maxTokens: number;
   temperature: number;
+  // setIsInThinkMode: (isInThinkMode: boolean) => void;
 };
 
 export enum MessageRole {
diff --git a/app-frontend/react/src/redux/Conversation/ConversationSlice.ts b/app-frontend/react/src/redux/Conversation/ConversationSlice.ts
index d93ae35..f695045 100644
--- a/app-frontend/react/src/redux/Conversation/ConversationSlice.ts
+++ b/app-frontend/react/src/redux/Conversation/ConversationSlice.ts
@@ -10,6 +10,7 @@ import { createAsyncThunkWrapper } from "../thunkUtil";
 import client from "../../common/client";
 import { notifications } from "@mantine/notifications";
 import { CHAT_QNA_URL, DATA_PREP_URL } from "../../config";
+// import { useState } from 'react';
 
 export interface FileDataSource {
   id: string;
@@ -172,14 +173,15 @@ export const isAgentSelector = (state: RootState) => state.conversationReducer.i
 
 export default ConversationSlice.reducer;
 
-let source: string[] = [];
-let content: any[] = [];
-let currentTool: string = "";
+// let source: string[] = [];
+// let content: any[] = [];
+// let currentTool: string = "";
 let isAgent: boolean = false;
 let currentAgentSteps: AgentStep[] = []; // Temporary storage for steps during streaming
 
 export const doConversation = (conversationRequest: ConversationRequest) => {
   const { conversationId, userPrompt, messages, model, maxTokens, temperature } = conversationRequest;
+  // const [isInThink, setIsInThink] = useState(false);
   if (!conversationId) {
     const id = uuidv4();
     store.dispatch(
@@ -193,7 +195,7 @@ export const doConversation = (conversationRequest: ConversationRequest) => {
   } else {
     store.dispatch(addMessageToMessages(userPrompt));
   }
-  
+
   const userPromptWithoutTime = {
     role: userPrompt.role,
     content: userPrompt.content,
@@ -206,17 +208,14 @@ export const doConversation = (conversationRequest: ConversationRequest) => {
     stream: true,
   };
 
-  function isJsonParsable(str: string): boolean {
-    try {
-      JSON.parse(str);
-      return true;
-    } catch (e) {
-      return false;
-    }
-  }
-
-  let result = "";
+  let result = ""; // Accumulates the final answer
+  let thinkBuffer = ""; // Accumulates data for think blocks
+  let postThinkBuffer = ""; // Accumulates plain text after last </think>
+  let isInThink = false; // Tracks if we're inside a <think> block
+  // setIsInThinkMode(false); // Reset the think mode state
   currentAgentSteps = []; // Reset steps for this message
+  isAgent = false; // Tracks if this is an agent message (set once, never reset)
+  let isMessageDispatched = false; // Tracks if the final message has been dispatched
 
   try {
     console.log("CHAT_QNA_URL", CHAT_QNA_URL);
@@ -239,59 +238,109 @@ export const doConversation = (conversationRequest: ConversationRequest) => {
         }
       },
       onmessage(msg) {
-        if (msg?.data !== "[DONE]") {
-          // console.log ( "check is json", isJsonParsable(msg.data) )
-          if (isJsonParsable(msg.data)) {
+        if (msg?.data === "[DONE]") {
+          // Stream is done, finalize the message
+          if (isAgent && thinkBuffer) {
+            processThinkContent(thinkBuffer);
+          }
+          if (!isMessageDispatched) {
+            // Use postThinkBuffer as the final answer if present
+            if (postThinkBuffer.trim()) {
+              result = postThinkBuffer.trim();
+            }
+            store.dispatch(setOnGoingResult(result));
+            store.dispatch(
+              addMessageToMessages({
+                role: MessageRole.Assistant,
+                content: result,
+                time: getCurrentTimeStamp(),
+                agentSteps: isAgent ? [...currentAgentSteps] : [],
+              }),
+            );
+            isMessageDispatched = true;
+          }
+          currentAgentSteps = []; // Clear steps for next message
+          postThinkBuffer = "";
+          return;
+        }
+
+        const data = msg?.data || "";
 
+        // Handle think blocks and non-think content
+        if (data.includes("<think>")) {
+          if (!isAgent) {
+            isAgent = true;
             store.dispatch(setIsAgent(true));
-            const currentMsg = JSON.parse(msg.data);
-            if (currentMsg.tool || currentMsg.source || currentMsg.content) {
-              currentAgentSteps.push({
-                tool: currentMsg.tool || currentTool,
-                content: currentMsg.content || [],
-                source: currentMsg.source || [],
-              });
-            }
-            currentTool = currentMsg.tool? currentMsg.tool: "";
-            source = currentMsg.source? currentMsg.source: "";
-            if (currentMsg.content) {
-              content = [...content, ...currentMsg.content];
-              result = currentMsg.content[0];
-            }
-            console.log(currentMsg);
-            console.log("currentTool", currentTool);
-            console.log("source", source);
-            console.log("content", content);
-          } else {
-            // isAgent 
-            try {
-              // const match = msg.data.match(/b'([^']*)'/);
-              // if (match && match[1] !== "</s>") {
-              //   const extractedText = match[1];
-              //   if (extractedText.includes("\\x")) {
-              //     const decodedText = decodeEscapedBytes(extractedText);
-              //     result += decodedText;
-              //   } else {
-              //     result += extractedText;
-              //   }
-              // } else if (!match) {
-              //   result += msg?.data;
-              // }
-              // store.dispatch(setIsAgent(false));
-
-              result += msg?.data;
-
-              if (result) {
+          }
+          // Split on <think> to handle content before it
+          const parts = data.split("<think>");
+          for (let i = 0; i < parts.length; i++) {
+            const part = parts[i];
+            if (i === 0 && !isInThink && part) {
+              // Content before <think> (non-think)
+              postThinkBuffer += part;
+              if (isAgent) {
+                store.dispatch(setOnGoingResult(postThinkBuffer));
+              } else {
+                result += part;
                 store.dispatch(setOnGoingResult(result));
               }
-            } catch (e) {
-              console.log("something wrong in msg", e);
-              throw e;
+            } else {
+              // Start or continue think block
+              isInThink = true;
+              // setIsInThinkMode(true); // Set think mode state
+              thinkBuffer += part;
+              // Check if part contains </think>
+              if (part.includes("</think>")) {
+                const [thinkContent, afterThink] = part.split("</think>", 2);
+                thinkBuffer = thinkBuffer.substring(0, thinkBuffer.indexOf(part)) + thinkContent;
+                processThinkContent(thinkBuffer);
+                thinkBuffer = "";
+                isInThink = false;
+                // setIsInThinkMode(false); // Reset think mode state
+                if (afterThink) {
+                  // Handle content after </think> as non-think
+                  if (!afterThink.includes("<think>")) {
+                    postThinkBuffer += afterThink;
+                    store.dispatch(setOnGoingResult(postThinkBuffer));
+                  } else {
+                    thinkBuffer = afterThink;
+                    isInThink = true;
+                    // setIsInThinkMode(true); // Set think mode state
+                  }
+                }
+              }
+            }
+          }
+        } else if (isInThink) {
+          // Accumulate within think block
+          thinkBuffer += data;
+          if (data.includes("</think>")) {
+            const [thinkContent, afterThink] = data.split("</think>", 2);
+            thinkBuffer = thinkBuffer.substring(0, thinkBuffer.lastIndexOf(data)) + thinkContent;
+            processThinkContent(thinkBuffer);
+            thinkBuffer = "";
+            isInThink = false;
+            // setIsInThinkMode(false); // Reset think mode state
+            if (afterThink) {
+              // Handle content after </think>
+              if (!afterThink.includes("<think>")) {
+                postThinkBuffer += afterThink;
+                store.dispatch(setOnGoingResult(postThinkBuffer));
+              } else {
+                thinkBuffer = afterThink;
+                isInThink = true;
+                // setIsInThinkMode(true); // Set think mode state
+              }
             }
           }
         } else {
+          // Non-agent or post-think plain text
           if (isAgent) {
-            console.log("final answer:", result);
+            postThinkBuffer += data;
+            store.dispatch(setOnGoingResult(postThinkBuffer));
+          } else {
+            result += data;
             store.dispatch(setOnGoingResult(result));
           }
         }
@@ -302,32 +351,130 @@ export const doConversation = (conversationRequest: ConversationRequest) => {
         throw err;
       },
       onclose() {
+        if (!isMessageDispatched && (result || postThinkBuffer || (isAgent && currentAgentSteps.length > 0))) {
+          // Use postThinkBuffer as the final answer if present
+          if (postThinkBuffer.trim()) {
+            result = postThinkBuffer.trim();
+          }
+          store.dispatch(setOnGoingResult(result));
+          store.dispatch(
+            addMessageToMessages({
+              role: MessageRole.Assistant,
+              content: result,
+              time: getCurrentTimeStamp(),
+              agentSteps: isAgent ? [...currentAgentSteps] : [],
+            }),
+          );
+          isMessageDispatched = true;
+        }
         store.dispatch(setOnGoingResult(""));
-        console.log("onclose", result);
-        store.dispatch(
-          addMessageToMessages({
-            role: MessageRole.Assistant,
-            content: result,
-            time: getCurrentTimeStamp(),
-            agentSteps: [...currentAgentSteps], // Store steps with this message
-          }),
-        );
-        currentAgentSteps = []; // Clear steps for the next message
-        // isAgent = false;
-        // store.dispatch(setIsAgent(false));
+        currentAgentSteps = [];
+        postThinkBuffer = "";
       },
     });
   } catch (err) {
     console.log(err);
   }
-};
 
-// function decodeEscapedBytes(str: string): string {
-//   const byteArray: number[] = str
-//     .split("\\x")
-//     .slice(1)
-//     .map((byte: string) => parseInt(byte, 16));
-//   return new TextDecoder("utf-8").decode(new Uint8Array(byteArray));
-// }
+  // Helper function to process content within <think> tags
+  function processThinkContent(content: string) {
+    content = content.trim();
+    if (!content) return;
+
+    const toolCallRegex = /TOOL CALL: (\{.*?\})/g;
+    const finalAnswerRegex = /FINAL ANSWER: (\{.*?\})/;
+    let stepContent: string[] = []; // Collect all reasoning for this think block
+    let tool: string = "reasoning"; // Default tool
+    let source: string[] = []; // Tool output
+
+    // Split content by final answer (if present)
+    let remainingContent = content;
+    const finalAnswerMatch = content.match(finalAnswerRegex);
+    if (finalAnswerMatch) {
+      try {
+        const finalAnswer = JSON.parse(finalAnswerMatch[1].replace("FINAL ANSWER: ", ""));
+        if (finalAnswer.answer) {
+          result = finalAnswer.answer;
+        }
+        remainingContent = content.split(finalAnswerMatch[0])[0].trim(); // Content before FINAL ANSWER
+        tool = "final_answer";
+      } catch (e) {
+        console.error("Error parsing final answer:", finalAnswerMatch[1], e);
+      }
+    }
+
+    // Process tool calls within the remaining content
+    const toolMatches = remainingContent.match(toolCallRegex) || [];
+    let currentContent = remainingContent;
+
+    if (toolMatches.length > 0) {
+      // Handle content before and after tool calls
+      toolMatches.forEach((toolCallStr) => {
+        const [beforeTool, afterTool] = currentContent.split(toolCallStr, 2);
+        if (beforeTool.trim()) {
+          stepContent.push(beforeTool.trim());
+        }
+
+        try {
+          // Attempt to parse the tool call JSON
+          let toolCall;
+          try {
+            toolCall = JSON.parse(toolCallStr.replace("TOOL CALL: ", ""));
+          } catch (e) {
+            console.error("Error parsing tool call JSON, attempting recovery:", toolCallStr, e);
+            // Attempt to extract tool and content manually
+            const toolMatch = toolCallStr.match(/"tool":\s*"([^"]+)"/);
+            const contentMatch = toolCallStr.match(/"tool_content":\s*\["([^"]+)"\]/);
+            toolCall = {
+              tool: toolMatch ? toolMatch[1] : "unknown",
+              args: {
+                tool_content: contentMatch ? [contentMatch[1]] : [],
+              },
+            };
+          }
+
+          tool = toolCall.tool || tool;
+          source = toolCall.args?.tool_content || source;
+
+          // Clean up afterTool to remove invalid JSON fragments
+          if (afterTool.trim()) {
+            // Remove any trailing malformed JSON (e.g., "Chinook?"}})
+            const cleanAfterTool = afterTool.replace(/[\s\S]*?(\}\s*)$/, "").trim();
+            if (cleanAfterTool) {
+              stepContent.push(cleanAfterTool);
+            }
+          }
+
+        } catch (e) {
+          console.error("Failed to process tool call:", toolCallStr, e);
+          stepContent.push(`[Error parsing tool call: ${toolCallStr}]`);
+        }
+
+        currentContent = afterTool;
+      });
+    } else {
+      // No tool calls, treat as reasoning
+      if (remainingContent.trim()) {
+        stepContent.push(remainingContent.trim());
+      }
+    }
+
+    // Add the step for this think block
+    if (stepContent.length > 0 || source.length > 0) {
+      currentAgentSteps.push({
+        tool,
+        content: stepContent,
+        source,
+      });
+    }
+
+    // Update onGoingResult to trigger UI update with latest steps
+    if (isAgent) {
+      const latestContent = currentAgentSteps.flatMap(step => step.content).join(" ");
+      const latestSource = source.length > 0 ? source.join(" ") : "";
+      store.dispatch(setOnGoingResult(latestContent + (latestSource ? " " + latestSource : "") + (postThinkBuffer ? " " + postThinkBuffer : "")));
+    }
+  }
+};
 
 export const getCurrentAgentSteps = () => currentAgentSteps; // Export for use in Conversation.tsx
\ No newline at end of file
diff --git a/setup-scripts/setup-genai-studio/playbooks/deploy-studio.yml b/setup-scripts/setup-genai-studio/playbooks/deploy-studio.yml
index 7e3683e..b6ba011 100644
--- a/setup-scripts/setup-genai-studio/playbooks/deploy-studio.yml
+++ b/setup-scripts/setup-genai-studio/playbooks/deploy-studio.yml
@@ -62,7 +62,7 @@
         MYSQL_HOST: "{{ mysql_host }}"
 
     - name: Wait for all pods to be ready in studio namespace
-      shell: kubectl wait --for=condition=ready pod --all --namespace=studio --timeout=300s
+      shell: kubectl wait --for=condition=ready pod --all --namespace=studio --timeout=420s
       register: pod_ready_check
       failed_when: pod_ready_check.rc != 0
       changed_when: false
\ No newline at end of file
diff --git a/studio-backend/app/templates/app/app.compose.yaml b/studio-backend/app/templates/app/app.compose.yaml
index ce9bfa6..fb12458 100644
--- a/studio-backend/app/templates/app/app.compose.yaml
+++ b/studio-backend/app/templates/app/app.compose.yaml
@@ -48,6 +48,6 @@ app-nginx:
     - BACKEND_SERVICE_IP=${public_host_ip}
     - BACKEND_SERVICE_PORT=8888
     - DATAPREP_SERVICE_IP=${public_host_ip}
-    - DATAPREP_SERVICE_PORT=6007
+    - DATAPREP_SERVICE_PORT=${prepare_doc_redis_prep_0_port}
   ipc: host
   restart: always
\ No newline at end of file
diff --git a/studio-backend/app/templates/microsvc-composes/sql-agent.yaml b/studio-backend/app/templates/microsvc-composes/sql-agent.yaml
index eb1ce6b..6e84a58 100644
--- a/studio-backend/app/templates/microsvc-composes/sql-agent.yaml
+++ b/studio-backend/app/templates/microsvc-composes/sql-agent.yaml
@@ -3,8 +3,8 @@
   container_name: "{{endpoint}}"
   ports:
     - "{{port_key}}:9096"
-  volumes:
-    - ./agent-tools/:/home/user/tools/
+  # volumes:
+  #   - ./agent-tools/:/home/user/tools/
   ipc: host
   environment:
     ip_address: ${public_host_ip}
@@ -20,7 +20,6 @@
     temperature: "{{temperature}}"
     max_new_tokens: "{{maxNewToken}}"
     stream: false
-    tools: /home/user/tools/worker_agent_tools.yaml
     require_human_feedback: false
     llm_endpoint_url: "http://${public_host_ip}:{{llm_port}}"
     port: 9096
diff --git a/studio-backend/app/templates/microsvc-manifests/rag-agent.yaml b/studio-backend/app/templates/microsvc-manifests/rag-agent.yaml
index 9992610..64212fd 100644
--- a/studio-backend/app/templates/microsvc-manifests/rag-agent.yaml
+++ b/studio-backend/app/templates/microsvc-manifests/rag-agent.yaml
@@ -55,14 +55,14 @@ spec:
           command: ["/bin/sh", "-c"]
           args:
             - |
-              TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+              TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
               OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
               REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
               BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
               TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
               if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
               DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-              curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+              curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
               # Conditional wait for remote service based on llm_engine
               if [ "$llm_engine" = "tgi" ]; then
diff --git a/studio-backend/app/templates/microsvc-manifests/sql-agent.yaml b/studio-backend/app/templates/microsvc-manifests/sql-agent.yaml
index 098b095..21b3cf4 100644
--- a/studio-backend/app/templates/microsvc-manifests/sql-agent.yaml
+++ b/studio-backend/app/templates/microsvc-manifests/sql-agent.yaml
@@ -47,37 +47,37 @@ spec:
       labels:
         app: "{endpoint}"
     spec:
-      initContainers:
-        - name: agentqna-tools
-          image: curlimages/curl:latest
-          command: ["/bin/sh", "-c"]
-          args:
-            - |
-              TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
-              OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
-              REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
-              BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
-              TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
-              if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
-              DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-              curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+      # initContainers:
+      #   - name: agentqna-tools
+      #     image: curlimages/curl:latest
+      #     command: ["/bin/sh", "-c"]
+      #     args:
+      #       - |
+      #         TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
+      #         OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
+      #         REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
+      #         BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
+      #         TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
+      #         if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
+      #         DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
+      #         curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
-              # Conditional wait for remote service based on llm_engine
-              if [ "$llm_engine" = "tgi" ]; then
-                until nc -z -v -w30 ${llm_endpoint_url#http://} 80; do
-                  echo "Waiting for remote service...";
-                  sleep 5;
-                done
-              fi
-          envFrom:
-            - configMapRef:
-                name: config-{endpoint}
-          volumeMounts:
-            - name: agent-tools
-              mountPath: /home/user/tools/
-          securityContext:
-            runAsUser: 0
-            runAsGroup: 0
+      #         # Conditional wait for remote service based on llm_engine
+      #         if [ "$llm_engine" = "tgi" ]; then
+      #           until nc -z -v -w30 ${llm_endpoint_url#http://} 80; do
+      #             echo "Waiting for remote service...";
+      #             sleep 5;
+      #           done
+      #         fi
+      #     envFrom:
+      #       - configMapRef:
+      #           name: config-{endpoint}
+      #     volumeMounts:
+      #       - name: agent-tools
+      #         mountPath: /home/user/tools/
+      #     securityContext:
+      #       runAsUser: 0
+      #       runAsGroup: 0
       containers:
         - name: sql-agent-container
           image: ${REGISTRY}/agent:${TAG}
@@ -92,9 +92,9 @@ spec:
               python agent.py
           ports:
             - containerPort: 9096
-          volumeMounts:
-            - name: agent-tools
-              mountPath: /home/user/tools/
+          # volumeMounts:
+          #   - name: agent-tools
+          #     mountPath: /home/user/tools/
           envFrom:
             - configMapRef:
                 name: config-{endpoint}
diff --git a/studio-backend/app/templates/microsvc-manifests/supervisor-agent.yaml b/studio-backend/app/templates/microsvc-manifests/supervisor-agent.yaml
index 235e425..c7fc067 100644
--- a/studio-backend/app/templates/microsvc-manifests/supervisor-agent.yaml
+++ b/studio-backend/app/templates/microsvc-manifests/supervisor-agent.yaml
@@ -55,14 +55,14 @@ spec:
           command: ["/bin/sh", "-c"]
           args:
             - |
-              TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+              TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
               OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
               REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
               BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
               TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
               if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
               DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-              curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+              curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
               # Conditional wait for remote service based on llm_engine
               if [ "$llm_engine" = "tgi" ]; then
diff --git a/studio-backend/tests/exporter-groundtruth/gt_app-manifest-with-nginx.yaml b/studio-backend/tests/exporter-groundtruth/gt_app-manifest-with-nginx.yaml
index 8714947..828007d 100644
--- a/studio-backend/tests/exporter-groundtruth/gt_app-manifest-with-nginx.yaml
+++ b/studio-backend/tests/exporter-groundtruth/gt_app-manifest-with-nginx.yaml
@@ -797,14 +797,14 @@ spec:
         - -c
         args:
         - |
-          TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+          TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
           OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
           REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
           BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
           TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
           if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
           DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
           # Conditional wait for remote service based on llm_engine
           if [ "$llm_engine" = "tgi" ]; then
@@ -1139,14 +1139,14 @@ spec:
         - -c
         args:
         - |
-          TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+          TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
           OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
           REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
           BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
           TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
           if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
           DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
           # Conditional wait for remote service based on llm_engine
           if [ "$llm_engine" = "tgi" ]; then
@@ -1253,14 +1253,14 @@ spec:
         - -c
         args:
         - |
-          TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+          TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
           OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
           REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
           BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
           TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
           if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
           DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
           # Conditional wait for remote service based on llm_engine
           if [ "$llm_engine" = "tgi" ]; then
diff --git a/studio-backend/tests/exporter-groundtruth/gt_app-manifest.yaml b/studio-backend/tests/exporter-groundtruth/gt_app-manifest.yaml
index 92f2d46..24edd31 100644
--- a/studio-backend/tests/exporter-groundtruth/gt_app-manifest.yaml
+++ b/studio-backend/tests/exporter-groundtruth/gt_app-manifest.yaml
@@ -797,14 +797,14 @@ spec:
         - -c
         args:
         - |
-          TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+          TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
           OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
           REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
           BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
           TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
           if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
           DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
           # Conditional wait for remote service based on llm_engine
           if [ "$llm_engine" = "tgi" ]; then
@@ -1139,14 +1139,14 @@ spec:
         - -c
         args:
         - |
-          TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+          TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
           OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
           REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
           BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
           TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
           if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
           DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
           # Conditional wait for remote service based on llm_engine
           if [ "$llm_engine" = "tgi" ]; then
@@ -1253,14 +1253,14 @@ spec:
         - -c
         args:
         - |
-          TOOLS_GIT_URL="https://github.com/wanhakim/GenAIStudio/tree/main/app-backend/templates/tools"
+          TOOLS_GIT_URL="https://github.com/opea-project/GenAIStudio/tree/main/studio-backend/app/templates/tools"
           OWNER=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\1|')
           REPO=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/([^/]+)/([^/]+)/tree/([^/]+)/.*|\2|')
           BRANCH=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/([^/]+)/.*|\1|')
           TOOLS_DIR=$(echo ${TOOLS_GIT_URL} | sed -E 's|https://github.com/[^/]+/[^/]+/tree/[^/]+/(.*?)/?$|\1|')
           if [[ "${TOOLS_DIR: -1}" == "/" ]]; then TOOLS_DIR="${TOOLS_DIR%/}"; fi
           DOWNLOAD_URL="https://codeload.github.com/${OWNER}/${REPO}/tar.gz/${BRANCH}"
-          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=4 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
+          curl "${DOWNLOAD_URL}" | tar -xz --strip-components=5 -C /home/user/tools/ "${REPO}-${BRANCH}/${TOOLS_DIR}"
 
           # Conditional wait for remote service based on llm_engine
           if [ "$llm_engine" = "tgi" ]; then