diff --git a/Dockerfile b/Dockerfile
index 1b231ad62..7c518d411 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -196,6 +196,7 @@ COPY deepdoc deepdoc
 COPY rag rag
 COPY agent agent
 COPY graphrag graphrag
+COPY agentic_reasoning agentic_reasoning
 COPY pyproject.toml uv.lock ./
 
 COPY docker/service_conf.yaml.template ./conf/service_conf.yaml.template
diff --git a/agent/component/generate.py b/agent/component/generate.py
index ea12f9040..b9ed75c38 100644
--- a/agent/component/generate.py
+++ b/agent/component/generate.py
@@ -18,10 +18,10 @@ from functools import partial
 import pandas as pd
 from api.db import LLMType
 from api.db.services.conversation_service import structure_answer
-from api.db.services.dialog_service import message_fit_in
 from api.db.services.llm_service import LLMBundle
 from api import settings
 from agent.component.base import ComponentBase, ComponentParamBase
+from rag.prompts import message_fit_in
 
 
 class GenerateParam(ComponentParamBase):
diff --git a/agent/component/retrieval.py b/agent/component/retrieval.py
index 766035c7a..50c6c7963 100644
--- a/agent/component/retrieval.py
+++ b/agent/component/retrieval.py
@@ -19,11 +19,11 @@ from abc import ABC
 import pandas as pd
 
 from api.db import LLMType
-from api.db.services.dialog_service import label_question
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
 from api import settings
 from agent.component.base import ComponentBase, ComponentParamBase
+from rag.app.tag import label_question
 
 
 class RetrievalParam(ComponentParamBase):
diff --git a/agentic_reasoning/__init__.py b/agentic_reasoning/__init__.py
new file mode 100644
index 000000000..1422de46e
--- /dev/null
+++ b/agentic_reasoning/__init__.py
@@ -0,0 +1 @@
+from .deep_research import DeepResearcher as DeepResearcher
\ No newline at end of file
diff --git a/agentic_reasoning/deep_research.py b/agentic_reasoning/deep_research.py
new file mode 100644
index 000000000..f61c3a881
--- /dev/null
+++ b/agentic_reasoning/deep_research.py
@@ -0,0 +1,167 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import logging
+import re
+from functools import partial
+from agentic_reasoning.prompts import BEGIN_SEARCH_QUERY, BEGIN_SEARCH_RESULT, END_SEARCH_RESULT, MAX_SEARCH_LIMIT, \
+    END_SEARCH_QUERY, REASON_PROMPT, RELEVANT_EXTRACTION_PROMPT
+from api.db.services.llm_service import LLMBundle
+from rag.nlp import extract_between
+from rag.prompts import kb_prompt
+from rag.utils.tavily_conn import Tavily
+
+
+class DeepResearcher:
+    def __init__(self,
+                 chat_mdl: LLMBundle,
+                 prompt_config: dict,
+                 kb_retrieve: partial = None,
+                 kg_retrieve: partial = None
+                 ):
+        self.chat_mdl = chat_mdl
+        self.prompt_config = prompt_config
+        self._kb_retrieve = kb_retrieve
+        self._kg_retrieve = kg_retrieve
+
+    def thinking(self, chunk_info: dict, question: str):
+        def rm_query_tags(line):
+            pattern = re.escape(BEGIN_SEARCH_QUERY) + r"(.*?)" + re.escape(END_SEARCH_QUERY)
+            return re.sub(pattern, "", line)
+
+        def rm_result_tags(line):
+            pattern = re.escape(BEGIN_SEARCH_RESULT) + r"(.*?)" + re.escape(END_SEARCH_RESULT)
+            return re.sub(pattern, "", line)
+
+        executed_search_queries = []
+        msg_hisotry = [{"role": "user", "content": f'Question:\"{question}\"\n'}]
+        all_reasoning_steps = []
+        think = "<think>"
+        for ii in range(MAX_SEARCH_LIMIT + 1):
+            if ii == MAX_SEARCH_LIMIT - 1:
+                summary_think = f"\n{BEGIN_SEARCH_RESULT}\nThe maximum search limit is exceeded. You are not allowed to search.\n{END_SEARCH_RESULT}\n"
+                yield {"answer": think + summary_think + "</think>", "reference": {}, "audio_binary": None}
+                all_reasoning_steps.append(summary_think)
+                msg_hisotry.append({"role": "assistant", "content": summary_think})
+                break
+
+            query_think = ""
+            if msg_hisotry[-1]["role"] != "user":
+                msg_hisotry.append({"role": "user", "content": "Continues reasoning with the new information.\n"})
+            else:
+                msg_hisotry[-1]["content"] += "\n\nContinues reasoning with the new information.\n"
+            for ans in self.chat_mdl.chat_streamly(REASON_PROMPT, msg_hisotry, {"temperature": 0.7}):
+                ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
+                if not ans:
+                    continue
+                query_think = ans
+                yield {"answer": think + rm_query_tags(query_think) + "</think>", "reference": {}, "audio_binary": None}
+
+            think += rm_query_tags(query_think)
+            all_reasoning_steps.append(query_think)
+            queries = extract_between(query_think, BEGIN_SEARCH_QUERY, END_SEARCH_QUERY)
+            if not queries:
+                if ii > 0:
+                    break
+                queries = [question]
+
+            for search_query in queries:
+                logging.info(f"[THINK]Query: {ii}. {search_query}")
+                msg_hisotry.append({"role": "assistant", "content": search_query})
+                think += f"\n\n> {ii +1}. {search_query}\n\n"
+                yield {"answer": think + "</think>", "reference": {}, "audio_binary": None}
+
+                summary_think = ""
+                # The search query has been searched in previous steps.
+                if search_query in executed_search_queries:
+                    summary_think = f"\n{BEGIN_SEARCH_RESULT}\nYou have searched this query. Please refer to previous results.\n{END_SEARCH_RESULT}\n"
+                    yield {"answer": think + summary_think + "</think>", "reference": {}, "audio_binary": None}
+                    all_reasoning_steps.append(summary_think)
+                    msg_hisotry.append({"role": "user", "content": summary_think})
+                    think += summary_think
+                    continue
+
+                truncated_prev_reasoning = ""
+                for i, step in enumerate(all_reasoning_steps):
+                    truncated_prev_reasoning += f"Step {i + 1}: {step}\n\n"
+
+                prev_steps = truncated_prev_reasoning.split('\n\n')
+                if len(prev_steps) <= 5:
+                    truncated_prev_reasoning = '\n\n'.join(prev_steps)
+                else:
+                    truncated_prev_reasoning = ''
+                    for i, step in enumerate(prev_steps):
+                        if i == 0 or i >= len(prev_steps) - 4 or BEGIN_SEARCH_QUERY in step or BEGIN_SEARCH_RESULT in step:
+                            truncated_prev_reasoning += step + '\n\n'
+                        else:
+                            if truncated_prev_reasoning[-len('\n\n...\n\n'):] != '\n\n...\n\n':
+                                truncated_prev_reasoning += '...\n\n'
+                truncated_prev_reasoning = truncated_prev_reasoning.strip('\n')
+
+                # Retrieval procedure:
+                # 1. KB search
+                # 2. Web search (optional)
+                # 3. KG search (optional)
+                kbinfos = self._kb_retrieve(question=search_query) if self._kb_retrieve else {"chunks": [], "doc_aggs": []}
+
+                if self.prompt_config.get("tavily_api_key"):
+                    tav = Tavily(self.prompt_config["tavily_api_key"])
+                    tav_res = tav.retrieve_chunks(" ".join(search_query))
+                    kbinfos["chunks"].extend(tav_res["chunks"])
+                    kbinfos["doc_aggs"].extend(tav_res["doc_aggs"])
+                if self.prompt_config.get("use_kg") and self._kg_retrieve:
+                    ck = self._kg_retrieve(question=search_query)
+                    if ck["content_with_weight"]:
+                        kbinfos["chunks"].insert(0, ck)
+
+                # Merge chunk info for citations
+                if not chunk_info["chunks"]:
+                    for k in chunk_info.keys():
+                        chunk_info[k] = kbinfos[k]
+                else:
+                    cids = [c["chunk_id"] for c in chunk_info["chunks"]]
+                    for c in kbinfos["chunks"]:
+                        if c["chunk_id"] in cids:
+                            continue
+                        chunk_info["chunks"].append(c)
+                    dids = [d["doc_id"] for d in chunk_info["doc_aggs"]]
+                    for d in kbinfos["doc_aggs"]:
+                        if d["doc_id"] in dids:
+                            continue
+                        chunk_info["doc_aggs"].append(d)
+
+                think += "\n\n"
+                for ans in self.chat_mdl.chat_streamly(
+                        RELEVANT_EXTRACTION_PROMPT.format(
+                            prev_reasoning=truncated_prev_reasoning,
+                            search_query=search_query,
+                            document="\n".join(kb_prompt(kbinfos, 4096))
+                        ),
+                        [{"role": "user",
+                          "content": f'Now you should analyze each web page and find helpful information based on the current search query "{search_query}" and previous reasoning steps.'}],
+                        {"temperature": 0.7}):
+                    ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
+                    if not ans:
+                        continue
+                    summary_think = ans
+                    yield {"answer": think + rm_result_tags(summary_think) + "</think>", "reference": {}, "audio_binary": None}
+
+                all_reasoning_steps.append(summary_think)
+                msg_hisotry.append(
+                    {"role": "user", "content": f"\n\n{BEGIN_SEARCH_RESULT}{summary_think}{END_SEARCH_RESULT}\n\n"})
+                think += rm_result_tags(summary_think)
+                logging.info(f"[THINK]Summary: {ii}. {summary_think}")
+
+        yield think + "</think>"
diff --git a/agentic_reasoning/prompts.py b/agentic_reasoning/prompts.py
new file mode 100644
index 000000000..610409af1
--- /dev/null
+++ b/agentic_reasoning/prompts.py
@@ -0,0 +1,112 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+
+BEGIN_SEARCH_QUERY = "<|begin_search_query|>"
+END_SEARCH_QUERY = "<|end_search_query|>"
+BEGIN_SEARCH_RESULT = "<|begin_search_result|>"
+END_SEARCH_RESULT = "<|end_search_result|>"
+MAX_SEARCH_LIMIT = 6
+
+REASON_PROMPT = (
+        "You are a reasoning assistant with the ability to perform dataset searches to help "
+        "you answer the user's question accurately. You have special tools:\n\n"
+        f"- To perform a search: write {BEGIN_SEARCH_QUERY} your query here {END_SEARCH_QUERY}.\n"
+        f"Then, the system will search and analyze relevant content, then provide you with helpful information in the format {BEGIN_SEARCH_RESULT} ...search results... {END_SEARCH_RESULT}.\n\n"
+        f"You can repeat the search process multiple times if necessary. The maximum number of search attempts is limited to {MAX_SEARCH_LIMIT}.\n\n"
+        "Once you have all the information you need, continue your reasoning.\n\n"
+        "-- Example 1 --\n" ########################################
+        "Question: \"Are both the directors of Jaws and Casino Royale from the same country?\"\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Who is the director of Jaws?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nThe director of Jaws is Steven Spielberg...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information.\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Where is Steven Spielberg from?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nSteven Allan Spielberg is an American filmmaker...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Who is the director of Casino Royale?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nCasino Royale is a 2006 spy film directed by Martin Campbell...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Where is Martin Campbell from?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nMartin Campbell (born 24 October 1943) is a New Zealand film and television director...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\nIt's enough to answer the question\n"
+
+        "-- Example 2 --\n" #########################################
+        "Question: \"When was the founder of craigslist born?\"\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY}Who was the founder of craigslist?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nCraigslist was founded by Craig Newmark...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information.\n"
+        "Assistant:\n"
+        f"    {BEGIN_SEARCH_QUERY} When was Craig Newmark born?{END_SEARCH_QUERY}\n\n"
+        "User:\n"
+        f"    {BEGIN_SEARCH_RESULT}\nCraig Newmark was born on December 6, 1952...\n{END_SEARCH_RESULT}\n\n"
+        "Continues reasoning with the new information...\n\n"
+        "Assistant:\nIt's enough to answer the question\n"
+        "**Remember**:\n"
+        f"- You have a dataset to search, so you just provide a proper search query.\n"
+        f"- Use {BEGIN_SEARCH_QUERY} to request a dataset search and end with {END_SEARCH_QUERY}.\n"
+        "- The language of query MUST be as the same as 'Question' or 'search result'.\n"
+        "- When done searching, continue your reasoning.\n\n"
+        'Please answer the following question. You should think step by step to solve it.\n\n'
+    )
+
+RELEVANT_EXTRACTION_PROMPT = """**Task Instruction:**
+
+    You are tasked with reading and analyzing web pages based on the following inputs: **Previous Reasoning Steps**, **Current Search Query**, and **Searched Web Pages**. Your objective is to extract relevant and helpful information for **Current Search Query** from the **Searched Web Pages** and seamlessly integrate this information into the **Previous Reasoning Steps** to continue reasoning for the original question.
+
+    **Guidelines:**
+
+    1. **Analyze the Searched Web Pages:**
+    - Carefully review the content of each searched web page.
+    - Identify factual information that is relevant to the **Current Search Query** and can aid in the reasoning process for the original question.
+
+    2. **Extract Relevant Information:**
+    - Select the information from the Searched Web Pages that directly contributes to advancing the **Previous Reasoning Steps**.
+    - Ensure that the extracted information is accurate and relevant.
+
+    3. **Output Format:**
+    - **If the web pages provide helpful information for current search query:** Present the information beginning with `**Final Information**` as shown below.
+    - The language of query **MUST BE** as the same as 'Search Query' or 'Web Pages'.\n"
+    **Final Information**
+
+    [Helpful information]
+
+    - **If the web pages do not provide any helpful information for current search query:** Output the following text.
+
+    **Final Information**
+
+    No helpful information found.
+
+    **Inputs:**
+    - **Previous Reasoning Steps:**  
+    {prev_reasoning}
+
+    - **Current Search Query:**  
+    {search_query}
+
+    - **Searched Web Pages:**  
+    {document}
+
+    """
diff --git a/api/apps/api_app.py b/api/apps/api_app.py
index cb260b6b2..533cffaee 100644
--- a/api/apps/api_app.py
+++ b/api/apps/api_app.py
@@ -25,7 +25,7 @@ from api.db import FileType, LLMType, ParserType, FileSource
 from api.db.db_models import APIToken, Task, File
 from api.db.services import duplicate_name
 from api.db.services.api_service import APITokenService, API4ConversationService
-from api.db.services.dialog_service import DialogService, chat, keyword_extraction, label_question
+from api.db.services.dialog_service import DialogService, chat
 from api.db.services.document_service import DocumentService, doc_upload_and_parse
 from api.db.services.file2document_service import File2DocumentService
 from api.db.services.file_service import FileService
@@ -38,6 +38,8 @@ from api.utils.api_utils import server_error_response, get_data_error_result, ge
     generate_confirmation_token
 
 from api.utils.file_utils import filename_type, thumbnail
+from rag.app.tag import label_question
+from rag.prompts import keyword_extraction
 from rag.utils.storage_factory import STORAGE_IMPL
 
 from api.db.services.canvas_service import UserCanvasService
diff --git a/api/apps/chunk_app.py b/api/apps/chunk_app.py
index 0de3db753..aff1ba866 100644
--- a/api/apps/chunk_app.py
+++ b/api/apps/chunk_app.py
@@ -19,9 +19,10 @@ import json
 from flask import request
 from flask_login import login_required, current_user
 
-from api.db.services.dialog_service import keyword_extraction, label_question
 from rag.app.qa import rmPrefix, beAdoc
+from rag.app.tag import label_question
 from rag.nlp import search, rag_tokenizer
+from rag.prompts import keyword_extraction
 from rag.settings import PAGERANK_FLD
 from rag.utils import rmSpace
 from api.db import LLMType, ParserType
diff --git a/api/apps/conversation_app.py b/api/apps/conversation_app.py
index 7abe52a4d..eba943a47 100644
--- a/api/apps/conversation_app.py
+++ b/api/apps/conversation_app.py
@@ -25,13 +25,14 @@ from flask import request, Response
 from flask_login import login_required, current_user
 
 from api.db import LLMType
-from api.db.services.dialog_service import DialogService, chat, ask, label_question
+from api.db.services.dialog_service import DialogService, chat, ask
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle, TenantService
 from api import settings
 from api.utils.api_utils import get_json_result
 from api.utils.api_utils import server_error_response, get_data_error_result, validate_request
 from graphrag.general.mind_map_extractor import MindMapExtractor
+from rag.app.tag import label_question
 
 
 @manager.route('/set', methods=['POST'])  # noqa: F821
diff --git a/api/apps/llm_app.py b/api/apps/llm_app.py
index 0ccb468bd..4a1ff183a 100644
--- a/api/apps/llm_app.py
+++ b/api/apps/llm_app.py
@@ -152,6 +152,7 @@ def add_llm():
 
     elif factory == "Tencent Cloud":
         req["api_key"] = apikey_json(["tencent_cloud_sid", "tencent_cloud_sk"])
+        return set_api_key()
 
     elif factory == "Bedrock":
         # For Bedrock, due to its special authentication method
diff --git a/api/apps/sdk/dify_retrieval.py b/api/apps/sdk/dify_retrieval.py
index 166f97049..5d6a8c896 100644
--- a/api/apps/sdk/dify_retrieval.py
+++ b/api/apps/sdk/dify_retrieval.py
@@ -16,11 +16,11 @@
 from flask import request, jsonify
 
 from api.db import LLMType
-from api.db.services.dialog_service import label_question
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import LLMBundle
 from api import settings
 from api.utils.api_utils import validate_request, build_error_result, apikey_required
+from rag.app.tag import label_question
 
 
 @manager.route('/dify/retrieval', methods=['POST'])  # noqa: F821
diff --git a/api/apps/sdk/doc.py b/api/apps/sdk/doc.py
index 98e2c4227..db3242251 100644
--- a/api/apps/sdk/doc.py
+++ b/api/apps/sdk/doc.py
@@ -16,7 +16,6 @@
 import pathlib
 import datetime
 
-from api.db.services.dialog_service import keyword_extraction, label_question
 from rag.app.qa import rmPrefix, beAdoc
 from rag.nlp import rag_tokenizer
 from api.db import LLMType, ParserType
@@ -39,6 +38,8 @@ from api.db.services.file_service import FileService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.utils.api_utils import construct_json_result, get_parser_config
 from rag.nlp import search
+from rag.prompts import keyword_extraction
+from rag.app.tag import label_question
 from rag.utils import rmSpace
 from rag.utils.storage_factory import STORAGE_IMPL
 
diff --git a/api/db/services/dialog_service.py b/api/db/services/dialog_service.py
index 0a4ab0f0e..6126ef150 100644
--- a/api/db/services/dialog_service.py
+++ b/api/db/services/dialog_service.py
@@ -15,31 +15,23 @@
 #
 import logging
 import binascii
-import os
-import json
 import time
-
-import json_repair
+from functools import partial
 import re
-from collections import defaultdict
 from copy import deepcopy
 from timeit import default_timer as timer
-import datetime
-from datetime import timedelta
+from agentic_reasoning import DeepResearcher
 from api.db import LLMType, ParserType, StatusEnum
 from api.db.db_models import Dialog, DB
 from api.db.services.common_service import CommonService
-from api.db.services.document_service import DocumentService
 from api.db.services.knowledgebase_service import KnowledgebaseService
 from api.db.services.llm_service import TenantLLMService, LLMBundle
 from api import settings
-from graphrag.utils import get_tags_from_cache, set_tags_to_cache
 from rag.app.resume import forbidden_select_fields4resume
-from rag.nlp import extract_between
+from rag.app.tag import label_question
 from rag.nlp.search import index_name
-from rag.settings import TAG_FLD
-from rag.utils import rmSpace, num_tokens_from_string, encoder
-from api.utils.file_utils import get_project_base_directory
+from rag.prompts import kb_prompt, message_fit_in, llm_id2llm_type, keyword_extraction, full_question
+from rag.utils import rmSpace, num_tokens_from_string
 from rag.utils.tavily_conn import Tavily
 
 
@@ -69,109 +61,6 @@ class DialogService(CommonService):
         return list(chats.dicts())
 
 
-def message_fit_in(msg, max_length=4000):
-    def count():
-        nonlocal msg
-        tks_cnts = []
-        for m in msg:
-            tks_cnts.append(
-                {"role": m["role"], "count": num_tokens_from_string(m["content"])})
-        total = 0
-        for m in tks_cnts:
-            total += m["count"]
-        return total
-
-    c = count()
-    if c < max_length:
-        return c, msg
-
-    msg_ = [m for m in msg[:-1] if m["role"] == "system"]
-    if len(msg) > 1:
-        msg_.append(msg[-1])
-    msg = msg_
-    c = count()
-    if c < max_length:
-        return c, msg
-
-    ll = num_tokens_from_string(msg_[0]["content"])
-    ll2 = num_tokens_from_string(msg_[-1]["content"])
-    if ll / (ll + ll2) > 0.8:
-        m = msg_[0]["content"]
-        m = encoder.decode(encoder.encode(m)[:max_length - ll2])
-        msg[0]["content"] = m
-        return max_length, msg
-
-    m = msg_[1]["content"]
-    m = encoder.decode(encoder.encode(m)[:max_length - ll2])
-    msg[1]["content"] = m
-    return max_length, msg
-
-
-def llm_id2llm_type(llm_id):
-    llm_id, _ = TenantLLMService.split_model_name_and_factory(llm_id)
-    fnm = os.path.join(get_project_base_directory(), "conf")
-    llm_factories = json.load(open(os.path.join(fnm, "llm_factories.json"), "r"))
-    for llm_factory in llm_factories["factory_llm_infos"]:
-        for llm in llm_factory["llm"]:
-            if llm_id == llm["llm_name"]:
-                return llm["model_type"].strip(",")[-1]
-
-
-def kb_prompt(kbinfos, max_tokens):
-    knowledges = [ck["content_with_weight"] for ck in kbinfos["chunks"]]
-    used_token_count = 0
-    chunks_num = 0
-    for i, c in enumerate(knowledges):
-        used_token_count += num_tokens_from_string(c)
-        chunks_num += 1
-        if max_tokens * 0.97 < used_token_count:
-            knowledges = knowledges[:i]
-            logging.warning(f"Not all the retrieval into prompt: {i+1}/{len(knowledges)}")
-            break
-
-    docs = DocumentService.get_by_ids([ck["doc_id"] for ck in kbinfos["chunks"][:chunks_num]])
-    docs = {d.id: d.meta_fields for d in docs}
-
-    doc2chunks = defaultdict(lambda: {"chunks": [], "meta": []})
-    for ck in kbinfos["chunks"][:chunks_num]:
-        doc2chunks[ck["docnm_kwd"]]["chunks"].append((f"URL: {ck['url']}\n" if "url" in ck else "") + ck["content_with_weight"])
-        doc2chunks[ck["docnm_kwd"]]["meta"] = docs.get(ck["doc_id"], {})
-
-    knowledges = []
-    for nm, cks_meta in doc2chunks.items():
-        txt = f"Document: {nm} \n"
-        for k, v in cks_meta["meta"].items():
-            txt += f"{k}: {v}\n"
-        txt += "Relevant fragments as following:\n"
-        for i, chunk in enumerate(cks_meta["chunks"], 1):
-            txt += f"{i}. {chunk}\n"
-        knowledges.append(txt)
-    return knowledges
-
-
-def label_question(question, kbs):
-    tags = None
-    tag_kb_ids = []
-    for kb in kbs:
-        if kb.parser_config.get("tag_kb_ids"):
-            tag_kb_ids.extend(kb.parser_config["tag_kb_ids"])
-    if tag_kb_ids:
-        all_tags = get_tags_from_cache(tag_kb_ids)
-        if not all_tags:
-            all_tags = settings.retrievaler.all_tags_in_portion(kb.tenant_id, tag_kb_ids)
-            set_tags_to_cache(all_tags, tag_kb_ids)
-        else:
-            all_tags = json.loads(all_tags)
-        tag_kbs = KnowledgebaseService.get_by_ids(tag_kb_ids)
-        tags = settings.retrievaler.tag_query(question,
-                                              list(set([kb.tenant_id for kb in tag_kbs])),
-                                              tag_kb_ids,
-                                              all_tags,
-                                              kb.parser_config.get("topn_tags", 3)
-                                              )
-    return tags
-
-
 def chat_solo(dialog, messages, stream=True):
     if llm_id2llm_type(dialog.llm_id) == "image2text":
         chat_mdl = LLMBundle(dialog.tenant_id, LLMType.IMAGE2TEXT, dialog.llm_id)
@@ -297,7 +186,11 @@ def chat(dialog, messages, stream=True, **kwargs):
 
         knowledges = []
         if prompt_config.get("reasoning", False):
-            for think in reasoning(kbinfos, " ".join(questions), chat_mdl, embd_mdl, tenant_ids, dialog.kb_ids, prompt_config, MAX_SEARCH_LIMIT=3):
+            reasoner = DeepResearcher(chat_mdl,
+                                      prompt_config,
+                                      partial(retriever.retrieval, embd_mdl=embd_mdl, tenant_ids=tenant_ids, kb_ids=dialog.kb_ids, page=1, page_size=dialog.top_n, similarity_threshold=0.2, vector_similarity_weight=0.3))
+
+            for think in reasoner.thinking(kbinfos, " ".join(questions)):
                 if isinstance(think, str):
                     thought = think
                     knowledges = [t for t in think.split("\n") if t]
@@ -552,175 +445,6 @@ Please write the SQL, only SQL, without any other explanations or text.
     }
 
 
-def relevant(tenant_id, llm_id, question, contents: list):
-    if llm_id2llm_type(llm_id) == "image2text":
-        chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
-    else:
-        chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
-    prompt = """
-        You are a grader assessing relevance of a retrieved document to a user question. 
-        It does not need to be a stringent test. The goal is to filter out erroneous retrievals.
-        If the document contains keyword(s) or semantic meaning related to the user question, grade it as relevant. 
-        Give a binary score 'yes' or 'no' score to indicate whether the document is relevant to the question.
-        No other words needed except 'yes' or 'no'.
-    """
-    if not contents:
-        return False
-    contents = "Documents: \n" + "   - ".join(contents)
-    contents = f"Question: {question}\n" + contents
-    if num_tokens_from_string(contents) >= chat_mdl.max_length - 4:
-        contents = encoder.decode(encoder.encode(contents)[:chat_mdl.max_length - 4])
-    ans = chat_mdl.chat(prompt, [{"role": "user", "content": contents}], {"temperature": 0.01})
-    if ans.lower().find("yes") >= 0:
-        return True
-    return False
-
-
-def rewrite(tenant_id, llm_id, question):
-    if llm_id2llm_type(llm_id) == "image2text":
-        chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
-    else:
-        chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
-    prompt = """
-        You are an expert at query expansion to generate a paraphrasing of a question.
-        I can't retrieval relevant information from the knowledge base by using user's question directly.     
-        You need to expand or paraphrase user's question by multiple ways such as using synonyms words/phrase, 
-        writing the abbreviation in its entirety, adding some extra descriptions or explanations, 
-        changing the way of expression, translating the original question into another language (English/Chinese), etc. 
-        And return 5 versions of question and one is from translation.
-        Just list the question. No other words are needed.
-    """
-    ans = chat_mdl.chat(prompt, [{"role": "user", "content": question}], {"temperature": 0.8})
-    return ans
-
-
-def keyword_extraction(chat_mdl, content, topn=3):
-    prompt = f"""
-Role: You're a text analyzer. 
-Task: extract the most important keywords/phrases of a given piece of text content.
-Requirements: 
-  - Summarize the text content, and give top {topn} important keywords/phrases.
-  - The keywords MUST be in language of the given piece of text content.
-  - The keywords are delimited by ENGLISH COMMA.
-  - Keywords ONLY in output.
-
-### Text Content 
-{content}
-
-"""
-    msg = [
-        {"role": "system", "content": prompt},
-        {"role": "user", "content": "Output: "}
-    ]
-    _, msg = message_fit_in(msg, chat_mdl.max_length)
-    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.2})
-    if isinstance(kwd, tuple):
-        kwd = kwd[0]
-    kwd = re.sub(r"<think>.*</think>", "", kwd, flags=re.DOTALL)
-    if kwd.find("**ERROR**") >= 0:
-        return ""
-    return kwd
-
-
-def question_proposal(chat_mdl, content, topn=3):
-    prompt = f"""
-Role: You're a text analyzer. 
-Task:  propose {topn} questions about a given piece of text content.
-Requirements: 
-  - Understand and summarize the text content, and propose top {topn} important questions.
-  - The questions SHOULD NOT have overlapping meanings.
-  - The questions SHOULD cover the main content of the text as much as possible.
-  - The questions MUST be in language of the given piece of text content.
-  - One question per line.
-  - Question ONLY in output.
-
-### Text Content 
-{content}
-
-"""
-    msg = [
-        {"role": "system", "content": prompt},
-        {"role": "user", "content": "Output: "}
-    ]
-    _, msg = message_fit_in(msg, chat_mdl.max_length)
-    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.2})
-    if isinstance(kwd, tuple):
-        kwd = kwd[0]
-    kwd = re.sub(r"<think>.*</think>", "", kwd, flags=re.DOTALL)
-    if kwd.find("**ERROR**") >= 0:
-        return ""
-    return kwd
-
-
-def full_question(tenant_id, llm_id, messages):
-    if llm_id2llm_type(llm_id) == "image2text":
-        chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
-    else:
-        chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
-    conv = []
-    for m in messages:
-        if m["role"] not in ["user", "assistant"]:
-            continue
-        conv.append("{}: {}".format(m["role"].upper(), m["content"]))
-    conv = "\n".join(conv)
-    today = datetime.date.today().isoformat()
-    yesterday = (datetime.date.today() - timedelta(days=1)).isoformat()
-    tomorrow = (datetime.date.today() + timedelta(days=1)).isoformat()
-    prompt = f"""
-Role: A helpful assistant
-
-Task and steps: 
-    1. Generate a full user question that would follow the conversation.
-    2. If the user's question involves relative date, you need to convert it into absolute date based on the current date, which is {today}. For example: 'yesterday' would be converted to {yesterday}.
-    
-Requirements & Restrictions:
-  - Text generated MUST be in the same language of the original user's question.
-  - If the user's latest question is completely, don't do anything, just return the original question.
-  - DON'T generate anything except a refined question.
-
-######################
--Examples-
-######################
-
-# Example 1
-## Conversation
-USER: What is the name of Donald Trump's father?
-ASSISTANT:  Fred Trump.
-USER: And his mother?
-###############
-Output: What's the name of Donald Trump's mother?
-
-------------
-# Example 2
-## Conversation
-USER: What is the name of Donald Trump's father?
-ASSISTANT:  Fred Trump.
-USER: And his mother?
-ASSISTANT:  Mary Trump.
-User: What's her full name?
-###############
-Output: What's the full name of Donald Trump's mother Mary Trump?
-
-------------
-# Example 3
-## Conversation
-USER: What's the weather today in London?
-ASSISTANT:  Cloudy.
-USER: What's about tomorrow in Rochester?
-###############
-Output: What's the weather in Rochester on {tomorrow}?
-######################
-
-# Real Data
-## Conversation
-{conv}
-###############
-    """
-    ans = chat_mdl.chat(prompt, [{"role": "user", "content": "Output: "}], {"temperature": 0.2})
-    ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
-    return ans if ans.find("**ERROR**") < 0 else messages[-1]["content"]
-
-
 def tts(tts_mdl, text):
     if not tts_mdl or not text:
         return
@@ -796,298 +520,3 @@ def ask(question, kb_ids, tenant_id):
     yield decorate_answer(answer)
 
 
-def content_tagging(chat_mdl, content, all_tags, examples, topn=3):
-    prompt = f"""
-Role: You're a text analyzer. 
-
-Task: Tag (put on some labels) to a given piece of text content based on the examples and the entire tag set.
-
-Steps:: 
-  - Comprehend the tag/label set.
-  - Comprehend examples which all consist of both text content and assigned tags with relevance score in format of JSON.
-  - Summarize the text content, and tag it with top {topn} most relevant tags from the set of tag/label and the corresponding relevance score.
-
-Requirements
-  - The tags MUST be from the tag set.
-  - The output MUST be in JSON format only, the key is tag and the value is its relevance score.
-  - The relevance score must be range from 1 to 10.
-  - Keywords ONLY in output.
-
-# TAG SET
-{", ".join(all_tags)}
-
-"""
-    for i, ex in enumerate(examples):
-        prompt += """
-# Examples {}
-### Text Content
-{}
-
-Output:
-{}
-
-        """.format(i, ex["content"], json.dumps(ex[TAG_FLD], indent=2, ensure_ascii=False))
-
-    prompt += f"""
-# Real Data
-### Text Content
-{content}
-
-"""
-    msg = [
-        {"role": "system", "content": prompt},
-        {"role": "user", "content": "Output: "}
-    ]
-    _, msg = message_fit_in(msg, chat_mdl.max_length)
-    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.5})
-    if isinstance(kwd, tuple):
-        kwd = kwd[0]
-    kwd = re.sub(r"<think>.*</think>", "", kwd, flags=re.DOTALL)
-    if kwd.find("**ERROR**") >= 0:
-        raise Exception(kwd)
-
-    try:
-        return json_repair.loads(kwd)
-    except json_repair.JSONDecodeError:
-        try:
-            result = kwd.replace(prompt[:-1], '').replace('user', '').replace('model', '').strip()
-            result = '{' + result.split('{')[1].split('}')[0] + '}'
-            return json_repair.loads(result)
-        except Exception as e:
-            logging.exception(f"JSON parsing error: {result} -> {e}")
-            raise e
-
-
-def reasoning(chunk_info: dict, question: str, chat_mdl: LLMBundle, embd_mdl: LLMBundle,
-              tenant_ids: list[str], kb_ids: list[str], prompt_config, MAX_SEARCH_LIMIT: int = 6,
-              top_n: int = 5, similarity_threshold: float = 0.4, vector_similarity_weight: float = 0.3):
-    BEGIN_SEARCH_QUERY = "<|begin_search_query|>"
-    END_SEARCH_QUERY = "<|end_search_query|>"
-    BEGIN_SEARCH_RESULT = "<|begin_search_result|>"
-    END_SEARCH_RESULT = "<|end_search_result|>"
-
-    def rm_query_tags(line):
-        pattern = re.escape(BEGIN_SEARCH_QUERY) + r"(.*?)" + re.escape(END_SEARCH_QUERY)
-        return re.sub(pattern, "", line)
-
-    def rm_result_tags(line):
-        pattern = re.escape(BEGIN_SEARCH_RESULT) + r"(.*?)" + re.escape(END_SEARCH_RESULT)
-        return re.sub(pattern, "", line)
-
-    reason_prompt = (
-        "You are a reasoning assistant with the ability to perform dataset searches to help "
-        "you answer the user's question accurately. You have special tools:\n\n"
-        f"- To perform a search: write {BEGIN_SEARCH_QUERY} your query here {END_SEARCH_QUERY}.\n"
-        f"Then, the system will search and analyze relevant content, then provide you with helpful information in the format {BEGIN_SEARCH_RESULT} ...search results... {END_SEARCH_RESULT}.\n\n"
-        f"You can repeat the search process multiple times if necessary. The maximum number of search attempts is limited to {MAX_SEARCH_LIMIT}.\n\n"
-        "Once you have all the information you need, continue your reasoning.\n\n"
-        "-- Example 1 --\n" ########################################
-        "Question: \"Are both the directors of Jaws and Casino Royale from the same country?\"\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Who is the director of Jaws?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nThe director of Jaws is Steven Spielberg...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information.\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Where is Steven Spielberg from?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nSteven Allan Spielberg is an American filmmaker...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Who is the director of Casino Royale?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nCasino Royale is a 2006 spy film directed by Martin Campbell...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Where is Martin Campbell from?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nMartin Campbell (born 24 October 1943) is a New Zealand film and television director...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\nIt's enough to answer the question\n"
-        
-        "-- Example 2 --\n" #########################################
-        "Question: \"When was the founder of craigslist born?\"\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY}Who was the founder of craigslist?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nCraigslist was founded by Craig Newmark...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information.\n"
-        "Assistant:\n"
-        f"    {BEGIN_SEARCH_QUERY} When was Craig Newmark born?{END_SEARCH_QUERY}\n\n"
-        "User:\n"
-        f"    {BEGIN_SEARCH_RESULT}\nCraig Newmark was born on December 6, 1952...\n{END_SEARCH_RESULT}\n\n"
-        "Continues reasoning with the new information...\n\n"
-        "Assistant:\nIt's enough to answer the question\n"
-        "**Remember**:\n"
-        f"- You have a dataset to search, so you just provide a proper search query.\n"
-        f"- Use {BEGIN_SEARCH_QUERY} to request a dataset search and end with {END_SEARCH_QUERY}.\n"
-        "- The language of query MUST be as the same as 'Question' or 'search result'.\n"
-        "- When done searching, continue your reasoning.\n\n"
-        'Please answer the following question. You should think step by step to solve it.\n\n'
-    )
-
-    relevant_extraction_prompt = """**Task Instruction:**
-
-    You are tasked with reading and analyzing web pages based on the following inputs: **Previous Reasoning Steps**, **Current Search Query**, and **Searched Web Pages**. Your objective is to extract relevant and helpful information for **Current Search Query** from the **Searched Web Pages** and seamlessly integrate this information into the **Previous Reasoning Steps** to continue reasoning for the original question.
-
-    **Guidelines:**
-
-    1. **Analyze the Searched Web Pages:**
-    - Carefully review the content of each searched web page.
-    - Identify factual information that is relevant to the **Current Search Query** and can aid in the reasoning process for the original question.
-
-    2. **Extract Relevant Information:**
-    - Select the information from the Searched Web Pages that directly contributes to advancing the **Previous Reasoning Steps**.
-    - Ensure that the extracted information is accurate and relevant.
-
-    3. **Output Format:**
-    - **If the web pages provide helpful information for current search query:** Present the information beginning with `**Final Information**` as shown below.
-    - The language of query **MUST BE** as the same as 'Search Query' or 'Web Pages'.\n"
-    **Final Information**
-
-    [Helpful information]
-
-    - **If the web pages do not provide any helpful information for current search query:** Output the following text.
-
-    **Final Information**
-
-    No helpful information found.
-
-    **Inputs:**
-    - **Previous Reasoning Steps:**  
-    {prev_reasoning}
-
-    - **Current Search Query:**  
-    {search_query}
-
-    - **Searched Web Pages:**  
-    {document}
-
-    """
-
-    executed_search_queries = []
-    msg_hisotry = [{"role": "user", "content": f'Question:\"{question}\"\n'}]
-    all_reasoning_steps = []
-    think = "<think>"
-    for ii in range(MAX_SEARCH_LIMIT + 1):
-        if ii == MAX_SEARCH_LIMIT - 1:
-            summary_think = f"\n{BEGIN_SEARCH_RESULT}\nThe maximum search limit is exceeded. You are not allowed to search.\n{END_SEARCH_RESULT}\n"
-            yield {"answer": think + summary_think + "</think>", "reference": {}, "audio_binary": None}
-            all_reasoning_steps.append(summary_think)
-            msg_hisotry.append({"role": "assistant", "content": summary_think})
-            break
-
-        query_think = ""
-        if msg_hisotry[-1]["role"] != "user":
-            msg_hisotry.append({"role": "user", "content": "Continues reasoning with the new information.\n"})
-        else:
-            msg_hisotry[-1]["content"] += "\n\nContinues reasoning with the new information.\n"
-        for ans in chat_mdl.chat_streamly(reason_prompt, msg_hisotry, {"temperature": 0.7}):
-            ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
-            if not ans:
-                continue
-            query_think = ans
-            yield {"answer": think + rm_query_tags(query_think) + "</think>", "reference": {}, "audio_binary": None}
-
-        think += rm_query_tags(query_think)
-        all_reasoning_steps.append(query_think)
-        queries = extract_between(query_think, BEGIN_SEARCH_QUERY, END_SEARCH_QUERY)
-        if not queries:
-            if ii > 0:
-                break
-            queries = [question]
-
-        for search_query in queries:
-            logging.info(f"[THINK]Query: {ii}. {search_query}")
-            msg_hisotry.append({"role": "assistant", "content": search_query})
-            think += f"\n\n> {ii+1}. {search_query}\n\n"
-            yield {"answer": think + "</think>", "reference": {}, "audio_binary": None}
-
-            summary_think = ""
-            # The search query has been searched in previous steps.
-            if search_query in executed_search_queries:
-                summary_think = f"\n{BEGIN_SEARCH_RESULT}\nYou have searched this query. Please refer to previous results.\n{END_SEARCH_RESULT}\n"
-                yield {"answer": think + summary_think + "</think>", "reference": {}, "audio_binary": None}
-                all_reasoning_steps.append(summary_think)
-                msg_hisotry.append({"role": "user", "content": summary_think})
-                think += summary_think
-                continue
-
-            truncated_prev_reasoning = ""
-            for i, step in enumerate(all_reasoning_steps):
-                truncated_prev_reasoning += f"Step {i + 1}: {step}\n\n"
-
-            prev_steps = truncated_prev_reasoning.split('\n\n')
-            if len(prev_steps) <= 5:
-                truncated_prev_reasoning = '\n\n'.join(prev_steps)
-            else:
-                truncated_prev_reasoning = ''
-                for i, step in enumerate(prev_steps):
-                    if i == 0 or i >= len(prev_steps) - 4 or BEGIN_SEARCH_QUERY in step or BEGIN_SEARCH_RESULT in step:
-                        truncated_prev_reasoning += step + '\n\n'
-                    else:
-                        if truncated_prev_reasoning[-len('\n\n...\n\n'):] != '\n\n...\n\n':
-                            truncated_prev_reasoning += '...\n\n'
-            truncated_prev_reasoning = truncated_prev_reasoning.strip('\n')
-
-            # Retrieval procedure:
-            # 1. KB search
-            # 2. Web search (optional)
-            # 3. KG search (optional)
-            kbinfos = settings.retrievaler.retrieval(search_query, embd_mdl, tenant_ids, kb_ids, 1, top_n,
-                                                     similarity_threshold,
-                                                     vector_similarity_weight
-                                                     )
-            if prompt_config.get("tavily_api_key", "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"):
-                tav = Tavily(prompt_config["tavily_api_key"])
-                tav_res = tav.retrieve_chunks(" ".join(search_query))
-                kbinfos["chunks"].extend(tav_res["chunks"])
-                kbinfos["doc_aggs"].extend(tav_res["doc_aggs"])
-            if prompt_config.get("use_kg"):
-                ck = settings.kg_retrievaler.retrieval(search_query,
-                                                       tenant_ids,
-                                                       kb_ids,
-                                                       embd_mdl,
-                                                       chat_mdl)
-                if ck["content_with_weight"]:
-                    kbinfos["chunks"].insert(0, ck)
-
-            # Merge chunk info for citations
-            if not chunk_info["chunks"]:
-                for k in chunk_info.keys():
-                    chunk_info[k] = kbinfos[k]
-            else:
-                cids = [c["chunk_id"] for c in chunk_info["chunks"]]
-                for c in kbinfos["chunks"]:
-                    if c["chunk_id"] in cids:
-                        continue
-                    chunk_info["chunks"].append(c)
-                dids = [d["doc_id"] for d in chunk_info["doc_aggs"]]
-                for d in kbinfos["doc_aggs"]:
-                    if d["doc_id"] in dids:
-                        continue
-                    chunk_info["doc_aggs"].append(d)
-
-            think += "\n\n"
-            for ans in chat_mdl.chat_streamly(
-                    relevant_extraction_prompt.format(
-                        prev_reasoning=truncated_prev_reasoning,
-                        search_query=search_query,
-                        document="\n".join(kb_prompt(kbinfos, 4096))
-                    ),
-                    [{"role": "user",
-                     "content": f'Now you should analyze each web page and find helpful information based on the current search query "{search_query}" and previous reasoning steps.'}],
-                    {"temperature": 0.7}):
-                ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
-                if not ans:
-                    continue
-                summary_think = ans
-                yield {"answer": think + rm_result_tags(summary_think) + "</think>", "reference": {}, "audio_binary": None}
-
-            all_reasoning_steps.append(summary_think)
-            msg_hisotry.append(
-                {"role": "user", "content": f"\n\n{BEGIN_SEARCH_RESULT}{summary_think}{END_SEARCH_RESULT}\n\n"})
-            think += rm_result_tags(summary_think)
-            logging.info(f"[THINK]Summary: {ii}. {summary_think}")
-
-    yield think + "</think>"
diff --git a/rag/app/tag.py b/rag/app/tag.py
index f5c29f2a8..7263bee5b 100644
--- a/rag/app/tag.py
+++ b/rag/app/tag.py
@@ -13,7 +13,7 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 #
-
+import json
 import re
 import csv
 from copy import deepcopy
@@ -121,6 +121,32 @@ def chunk(filename, binary=None, lang="Chinese", callback=None, **kwargs):
         "Excel, csv(txt) format files are supported.")
 
 
+def label_question(question, kbs):
+    from api.db.services.knowledgebase_service import KnowledgebaseService
+    from graphrag.utils import get_tags_from_cache, set_tags_to_cache
+    from api import settings
+    tags = None
+    tag_kb_ids = []
+    for kb in kbs:
+        if kb.parser_config.get("tag_kb_ids"):
+            tag_kb_ids.extend(kb.parser_config["tag_kb_ids"])
+    if tag_kb_ids:
+        all_tags = get_tags_from_cache(tag_kb_ids)
+        if not all_tags:
+            all_tags = settings.retrievaler.all_tags_in_portion(kb.tenant_id, tag_kb_ids)
+            set_tags_to_cache(all_tags, tag_kb_ids)
+        else:
+            all_tags = json.loads(all_tags)
+        tag_kbs = KnowledgebaseService.get_by_ids(tag_kb_ids)
+        tags = settings.retrievaler.tag_query(question,
+                                              list(set([kb.tenant_id for kb in tag_kbs])),
+                                              tag_kb_ids,
+                                              all_tags,
+                                              kb.parser_config.get("topn_tags", 3)
+                                              )
+    return tags
+
+
 if __name__ == "__main__":
     import sys
 
diff --git a/rag/nlp/query.py b/rag/nlp/query.py
index 345f8cffa..d3c2f1432 100644
--- a/rag/nlp/query.py
+++ b/rag/nlp/query.py
@@ -88,7 +88,7 @@ class FulltextQueryer:
             tks_w = [(re.sub(r"^[\+-]", "", tk), w) for tk, w in tks_w if tk]
             tks_w = [(tk.strip(), w) for tk, w in tks_w if tk.strip()]
             syns = []
-            for tk, w in tks_w:
+            for tk, w in tks_w[:256]:
                 syn = self.syn.lookup(tk)
                 syn = rag_tokenizer.tokenize(" ".join(syn)).split()
                 keywords.extend(syn)
diff --git a/rag/prompts.py b/rag/prompts.py
new file mode 100644
index 000000000..e8497fee3
--- /dev/null
+++ b/rag/prompts.py
@@ -0,0 +1,297 @@
+#
+#  Copyright 2024 The InfiniFlow Authors. All Rights Reserved.
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+#
+import datetime
+import json
+import logging
+import os
+import re
+from collections import defaultdict
+import json_repair
+from api.db import LLMType
+from api.db.services.document_service import DocumentService
+from api.db.services.llm_service import TenantLLMService, LLMBundle
+from api.utils.file_utils import get_project_base_directory
+from rag.settings import TAG_FLD
+from rag.utils import num_tokens_from_string, encoder
+
+
+def llm_id2llm_type(llm_id):
+    llm_id, _ = TenantLLMService.split_model_name_and_factory(llm_id)
+    fnm = os.path.join(get_project_base_directory(), "conf")
+    llm_factories = json.load(open(os.path.join(fnm, "llm_factories.json"), "r"))
+    for llm_factory in llm_factories["factory_llm_infos"]:
+        for llm in llm_factory["llm"]:
+            if llm_id == llm["llm_name"]:
+                return llm["model_type"].strip(",")[-1]
+
+
+def message_fit_in(msg, max_length=4000):
+    def count():
+        nonlocal msg
+        tks_cnts = []
+        for m in msg:
+            tks_cnts.append(
+                {"role": m["role"], "count": num_tokens_from_string(m["content"])})
+        total = 0
+        for m in tks_cnts:
+            total += m["count"]
+        return total
+
+    c = count()
+    if c < max_length:
+        return c, msg
+
+    msg_ = [m for m in msg[:-1] if m["role"] == "system"]
+    if len(msg) > 1:
+        msg_.append(msg[-1])
+    msg = msg_
+    c = count()
+    if c < max_length:
+        return c, msg
+
+    ll = num_tokens_from_string(msg_[0]["content"])
+    ll2 = num_tokens_from_string(msg_[-1]["content"])
+    if ll / (ll + ll2) > 0.8:
+        m = msg_[0]["content"]
+        m = encoder.decode(encoder.encode(m)[:max_length - ll2])
+        msg[0]["content"] = m
+        return max_length, msg
+
+    m = msg_[1]["content"]
+    m = encoder.decode(encoder.encode(m)[:max_length - ll2])
+    msg[1]["content"] = m
+    return max_length, msg
+
+
+def kb_prompt(kbinfos, max_tokens):
+    knowledges = [ck["content_with_weight"] for ck in kbinfos["chunks"]]
+    used_token_count = 0
+    chunks_num = 0
+    for i, c in enumerate(knowledges):
+        used_token_count += num_tokens_from_string(c)
+        chunks_num += 1
+        if max_tokens * 0.97 < used_token_count:
+            knowledges = knowledges[:i]
+            logging.warning(f"Not all the retrieval into prompt: {i+1}/{len(knowledges)}")
+            break
+
+    docs = DocumentService.get_by_ids([ck["doc_id"] for ck in kbinfos["chunks"][:chunks_num]])
+    docs = {d.id: d.meta_fields for d in docs}
+
+    doc2chunks = defaultdict(lambda: {"chunks": [], "meta": []})
+    for ck in kbinfos["chunks"][:chunks_num]:
+        doc2chunks[ck["docnm_kwd"]]["chunks"].append((f"URL: {ck['url']}\n" if "url" in ck else "") + ck["content_with_weight"])
+        doc2chunks[ck["docnm_kwd"]]["meta"] = docs.get(ck["doc_id"], {})
+
+    knowledges = []
+    for nm, cks_meta in doc2chunks.items():
+        txt = f"Document: {nm} \n"
+        for k, v in cks_meta["meta"].items():
+            txt += f"{k}: {v}\n"
+        txt += "Relevant fragments as following:\n"
+        for i, chunk in enumerate(cks_meta["chunks"], 1):
+            txt += f"{i}. {chunk}\n"
+        knowledges.append(txt)
+    return knowledges
+
+
+def keyword_extraction(chat_mdl, content, topn=3):
+    prompt = f"""
+Role: You're a text analyzer. 
+Task: extract the most important keywords/phrases of a given piece of text content.
+Requirements: 
+  - Summarize the text content, and give top {topn} important keywords/phrases.
+  - The keywords MUST be in language of the given piece of text content.
+  - The keywords are delimited by ENGLISH COMMA.
+  - Keywords ONLY in output.
+
+### Text Content 
+{content}
+
+"""
+    msg = [
+        {"role": "system", "content": prompt},
+        {"role": "user", "content": "Output: "}
+    ]
+    _, msg = message_fit_in(msg, chat_mdl.max_length)
+    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.2})
+    if isinstance(kwd, tuple):
+        kwd = kwd[0]
+    kwd = re.sub(r"<think>.*</think>", "", kwd, flags=re.DOTALL)
+    if kwd.find("**ERROR**") >= 0:
+        return ""
+    return kwd
+
+
+def question_proposal(chat_mdl, content, topn=3):
+    prompt = f"""
+Role: You're a text analyzer. 
+Task:  propose {topn} questions about a given piece of text content.
+Requirements: 
+  - Understand and summarize the text content, and propose top {topn} important questions.
+  - The questions SHOULD NOT have overlapping meanings.
+  - The questions SHOULD cover the main content of the text as much as possible.
+  - The questions MUST be in language of the given piece of text content.
+  - One question per line.
+  - Question ONLY in output.
+
+### Text Content 
+{content}
+
+"""
+    msg = [
+        {"role": "system", "content": prompt},
+        {"role": "user", "content": "Output: "}
+    ]
+    _, msg = message_fit_in(msg, chat_mdl.max_length)
+    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.2})
+    if isinstance(kwd, tuple):
+        kwd = kwd[0]
+    kwd = re.sub(r"<think>.*</think>", "", kwd, flags=re.DOTALL)
+    if kwd.find("**ERROR**") >= 0:
+        return ""
+    return kwd
+
+
+def full_question(tenant_id, llm_id, messages):
+    if llm_id2llm_type(llm_id) == "image2text":
+        chat_mdl = LLMBundle(tenant_id, LLMType.IMAGE2TEXT, llm_id)
+    else:
+        chat_mdl = LLMBundle(tenant_id, LLMType.CHAT, llm_id)
+    conv = []
+    for m in messages:
+        if m["role"] not in ["user", "assistant"]:
+            continue
+        conv.append("{}: {}".format(m["role"].upper(), m["content"]))
+    conv = "\n".join(conv)
+    today = datetime.date.today().isoformat()
+    yesterday = (datetime.date.today() - datetime.timedelta(days=1)).isoformat()
+    tomorrow = (datetime.date.today() + datetime.timedelta(days=1)).isoformat()
+    prompt = f"""
+Role: A helpful assistant
+
+Task and steps: 
+    1. Generate a full user question that would follow the conversation.
+    2. If the user's question involves relative date, you need to convert it into absolute date based on the current date, which is {today}. For example: 'yesterday' would be converted to {yesterday}.
+
+Requirements & Restrictions:
+  - Text generated MUST be in the same language of the original user's question.
+  - If the user's latest question is completely, don't do anything, just return the original question.
+  - DON'T generate anything except a refined question.
+
+######################
+-Examples-
+######################
+
+# Example 1
+## Conversation
+USER: What is the name of Donald Trump's father?
+ASSISTANT:  Fred Trump.
+USER: And his mother?
+###############
+Output: What's the name of Donald Trump's mother?
+
+------------
+# Example 2
+## Conversation
+USER: What is the name of Donald Trump's father?
+ASSISTANT:  Fred Trump.
+USER: And his mother?
+ASSISTANT:  Mary Trump.
+User: What's her full name?
+###############
+Output: What's the full name of Donald Trump's mother Mary Trump?
+
+------------
+# Example 3
+## Conversation
+USER: What's the weather today in London?
+ASSISTANT:  Cloudy.
+USER: What's about tomorrow in Rochester?
+###############
+Output: What's the weather in Rochester on {tomorrow}?
+######################
+
+# Real Data
+## Conversation
+{conv}
+###############
+    """
+    ans = chat_mdl.chat(prompt, [{"role": "user", "content": "Output: "}], {"temperature": 0.2})
+    ans = re.sub(r"<think>.*</think>", "", ans, flags=re.DOTALL)
+    return ans if ans.find("**ERROR**") < 0 else messages[-1]["content"]
+
+
+def content_tagging(chat_mdl, content, all_tags, examples, topn=3):
+    prompt = f"""
+Role: You're a text analyzer. 
+
+Task: Tag (put on some labels) to a given piece of text content based on the examples and the entire tag set.
+
+Steps:: 
+  - Comprehend the tag/label set.
+  - Comprehend examples which all consist of both text content and assigned tags with relevance score in format of JSON.
+  - Summarize the text content, and tag it with top {topn} most relevant tags from the set of tag/label and the corresponding relevance score.
+
+Requirements
+  - The tags MUST be from the tag set.
+  - The output MUST be in JSON format only, the key is tag and the value is its relevance score.
+  - The relevance score must be range from 1 to 10.
+  - Keywords ONLY in output.
+
+# TAG SET
+{", ".join(all_tags)}
+
+"""
+    for i, ex in enumerate(examples):
+        prompt += """
+# Examples {}
+### Text Content
+{}
+
+Output:
+{}
+
+        """.format(i, ex["content"], json.dumps(ex[TAG_FLD], indent=2, ensure_ascii=False))
+
+    prompt += f"""
+# Real Data
+### Text Content
+{content}
+
+"""
+    msg = [
+        {"role": "system", "content": prompt},
+        {"role": "user", "content": "Output: "}
+    ]
+    _, msg = message_fit_in(msg, chat_mdl.max_length)
+    kwd = chat_mdl.chat(prompt, msg[1:], {"temperature": 0.5})
+    if isinstance(kwd, tuple):
+        kwd = kwd[0]
+    kwd = re.sub(r"<think>.*</think>", "", kwd, flags=re.DOTALL)
+    if kwd.find("**ERROR**") >= 0:
+        raise Exception(kwd)
+
+    try:
+        return json_repair.loads(kwd)
+    except json_repair.JSONDecodeError:
+        try:
+            result = kwd.replace(prompt[:-1], '').replace('user', '').replace('model', '').strip()
+            result = '{' + result.split('{')[1].split('}')[0] + '}'
+            return json_repair.loads(result)
+        except Exception as e:
+            logging.exception(f"JSON parsing error: {result} -> {e}")
+            raise e
diff --git a/rag/svr/task_executor.py b/rag/svr/task_executor.py
index 952ed0825..0dd239742 100644
--- a/rag/svr/task_executor.py
+++ b/rag/svr/task_executor.py
@@ -23,6 +23,7 @@ from graphrag.general.index import WithCommunity, WithResolution, Dealer
 from graphrag.light.graph_extractor import GraphExtractor as LightKGExt
 from graphrag.general.graph_extractor import GraphExtractor as GeneralKGExt
 from graphrag.utils import get_llm_cache, set_llm_cache, get_tags_from_cache, set_tags_to_cache
+from rag.prompts import keyword_extraction, question_proposal, content_tagging
 
 CONSUMER_NO = "0" if len(sys.argv) < 2 else sys.argv[1]
 CONSUMER_NAME = "task_executor_" + CONSUMER_NO
@@ -49,7 +50,6 @@ import numpy as np
 from peewee import DoesNotExist
 
 from api.db import LLMType, ParserType, TaskStatus
-from api.db.services.dialog_service import keyword_extraction, question_proposal, content_tagging
 from api.db.services.document_service import DocumentService
 from api.db.services.llm_service import LLMBundle
 from api.db.services.task_service import TaskService
diff --git a/uv.lock b/uv.lock
index 850cde6e6..d0c120475 100644
--- a/uv.lock
+++ b/uv.lock
@@ -1084,6 +1084,10 @@ name = "datrie"
 version = "0.8.2"
 source = { registry = "https://mirrors.aliyun.com/pypi/simple" }
 sdist = { url = "https://mirrors.aliyun.com/pypi/packages/9d/fe/db74bd405d515f06657f11ad529878fd389576dca4812bea6f98d9b31574/datrie-0.8.2.tar.gz", hash = "sha256:525b08f638d5cf6115df6ccd818e5a01298cd230b2dac91c8ff2e6499d18765d" }
+wheels = [
+    { url = "https://mirrors.aliyun.com/pypi/packages/44/02/53f0cf0bf0cd629ba6c2cc13f2f9db24323459e9c19463783d890a540a96/datrie-0.8.2-pp273-pypy_73-win32.whl", hash = "sha256:b07bd5fdfc3399a6dab86d6e35c72b1dbd598e80c97509c7c7518ab8774d3fda" },
+]
+
 
 [[package]]
 name = "decorator"
@@ -4239,6 +4243,10 @@ wheels = [
     { url = "https://mirrors.aliyun.com/pypi/packages/48/7d/0f2b09490b98cc6a902ac15dda8760c568b9c18cfe70e0ef7a16de64d53a/pycryptodomex-3.20.0-cp35-abi3-musllinux_1_1_x86_64.whl", hash = "sha256:7a7a8f33a1f1fb762ede6cc9cbab8f2a9ba13b196bfaf7bc6f0b39d2ba315a43" },
     { url = "https://mirrors.aliyun.com/pypi/packages/b0/1c/375adb14b71ee1c8d8232904e928b3e7af5bbbca7c04e4bec94fe8e90c3d/pycryptodomex-3.20.0-cp35-abi3-win32.whl", hash = "sha256:c39778fd0548d78917b61f03c1fa8bfda6cfcf98c767decf360945fe6f97461e" },
     { url = "https://mirrors.aliyun.com/pypi/packages/b2/e8/1b92184ab7e5595bf38000587e6f8cf9556ebd1bf0a583619bee2057afbd/pycryptodomex-3.20.0-cp35-abi3-win_amd64.whl", hash = "sha256:2a47bcc478741b71273b917232f521fd5704ab4b25d301669879e7273d3586cc" },
+
+    { url = "https://mirrors.aliyun.com/pypi/packages/e7/c5/9140bb867141d948c8e242013ec8a8011172233c898dfdba0a2417c3169a/pycryptodomex-3.20.0-pp27-pypy_73-manylinux2010_x86_64.whl", hash = "sha256:1be97461c439a6af4fe1cf8bf6ca5936d3db252737d2f379cc6b2e394e12a458" },
+    { url = "https://mirrors.aliyun.com/pypi/packages/5e/6a/04acb4978ce08ab16890c70611ebc6efd251681341617bbb9e53356dee70/pycryptodomex-3.20.0-pp27-pypy_73-win32.whl", hash = "sha256:19764605feea0df966445d46533729b645033f134baeb3ea26ad518c9fdf212c" },
+
     { url = "https://mirrors.aliyun.com/pypi/packages/eb/df/3f1ea084e43b91e6d2b6b3493cc948864c17ea5d93ff1261a03812fbfd1a/pycryptodomex-3.20.0-pp310-pypy310_pp73-macosx_10_9_x86_64.whl", hash = "sha256:f2e497413560e03421484189a6b65e33fe800d3bd75590e6d78d4dfdb7accf3b" },
     { url = "https://mirrors.aliyun.com/pypi/packages/c9/f3/83ffbdfa0c8f9154bcd8866895f6cae5a3ec749da8b0840603cf936c4412/pycryptodomex-3.20.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:e48217c7901edd95f9f097feaa0388da215ed14ce2ece803d3f300b4e694abea" },
     { url = "https://mirrors.aliyun.com/pypi/packages/c9/9d/c113e640aaf02af5631ae2686b742aac5cd0e1402b9d6512b1c7ec5ef05d/pycryptodomex-3.20.0-pp310-pypy310_pp73-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:d00fe8596e1cc46b44bf3907354e9377aa030ec4cd04afbbf6e899fc1e2a7781" },