feat: config max_search_results for search engine (#192)

* feat: implement UI

* feat: config max_search_results for search engine via api

---------

Co-authored-by: Henry Li <henry1943@163.com>

src/agents/__init__.py

@@ -1,6 +1,6 @@
 # Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
 # SPDX-License-Identifier: MIT
 
-from .agents import research_agent, coder_agent
+from .agents import create_agent
 
-__all__ = ["research_agent", "coder_agent"]
+__all__ = ["create_agent"]

src/agents/agents.py

@@ -4,12 +4,6 @@
 from langgraph.prebuilt import create_react_agent
 
 from src.prompts import apply_prompt_template
-from src.tools import (
-    crawl_tool,
-    python_repl_tool,
-    web_search_tool,
-)
-
 from src.llms.llm import get_llm_by_type
 from src.config.agents import AGENT_LLM_MAP
 
@@ -23,10 +17,3 @@ def create_agent(agent_name: str, agent_type: str, tools: list, prompt_template:
         tools=tools,
         prompt=lambda state: apply_prompt_template(prompt_template, state),
     )
-
-
-# Create agents using the factory function
-research_agent = create_agent(
-    "researcher", "researcher", [web_search_tool, crawl_tool], "researcher"
-)
-coder_agent = create_agent("coder", "coder", [python_repl_tool], "coder")

src/config/__init__.py

@@ -1,7 +1,7 @@
 # Copyright (c) 2025 Bytedance Ltd. and/or its affiliates
 # SPDX-License-Identifier: MIT
 
-from .tools import SEARCH_MAX_RESULTS, SELECTED_SEARCH_ENGINE, SearchEngine
+from .tools import SELECTED_SEARCH_ENGINE, SearchEngine
 from .loader import load_yaml_config
 from .questions import BUILT_IN_QUESTIONS, BUILT_IN_QUESTIONS_ZH_CN
 
@@ -42,7 +42,6 @@ __all__ = [
     # Other configurations
     "TEAM_MEMBERS",
     "TEAM_MEMBER_CONFIGRATIONS",
-    "SEARCH_MAX_RESULTS",
     "SELECTED_SEARCH_ENGINE",
     "SearchEngine",
     "BUILT_IN_QUESTIONS",

src/config/configuration.py

@@ -14,6 +14,7 @@ class Configuration:
 
     max_plan_iterations: int = 1  # Maximum number of plan iterations
     max_step_num: int = 3  # Maximum number of steps in a plan
+    max_search_results: int = 3  # Maximum number of search results
     mcp_settings: dict = None  # MCP settings, including dynamic loaded tools
 
     @classmethod

src/config/tools.py

@@ -17,4 +17,3 @@ class SearchEngine(enum.Enum):
 
 # Tool configuration
 SELECTED_SEARCH_ENGINE = os.getenv("SEARCH_API", SearchEngine.TAVILY.value)
-SEARCH_MAX_RESULTS = 3

src/graph/nodes.py

@@ -12,12 +12,11 @@ from langchain_core.tools import tool
 from langgraph.types import Command, interrupt
 from langchain_mcp_adapters.client import MultiServerMCPClient
 
-from src.agents.agents import coder_agent, research_agent, create_agent
-
+from src.agents import create_agent
 from src.tools.search import LoggedTavilySearch
 from src.tools import (
     crawl_tool,
-    web_search_tool,
+    get_web_search_tool,
     python_repl_tool,
 )
 
@@ -29,7 +28,7 @@ from src.prompts.template import apply_prompt_template
 from src.utils.json_utils import repair_json_output
 
 from .types import State
-from ..config import SEARCH_MAX_RESULTS, SELECTED_SEARCH_ENGINE, SearchEngine
+from ..config import SELECTED_SEARCH_ENGINE, SearchEngine
 
 logger = logging.getLogger(__name__)
 
@@ -45,13 +44,16 @@ def handoff_to_planner(
     return
 
 
-def background_investigation_node(state: State) -> Command[Literal["planner"]]:
+def background_investigation_node(
+    state: State, config: RunnableConfig
+) -> Command[Literal["planner"]]:
     logger.info("background investigation node is running.")
+    configurable = Configuration.from_runnable_config(config)
     query = state["messages"][-1].content
     if SELECTED_SEARCH_ENGINE == SearchEngine.TAVILY:
-        searched_content = LoggedTavilySearch(max_results=SEARCH_MAX_RESULTS).invoke(
-            {"query": query}
-        )
+        searched_content = LoggedTavilySearch(
+            max_results=configurable.max_search_results
+        ).invoke({"query": query})
         background_investigation_results = None
         if isinstance(searched_content, list):
             background_investigation_results = [
@@ -63,7 +65,9 @@ def background_investigation_node(state: State) -> Command[Literal["planner"]]:
                 f"Tavily search returned malformed response: {searched_content}"
             )
     else:
-        background_investigation_results = web_search_tool.invoke(query)
+        background_investigation_results = get_web_search_tool(
+            configurable.max_search_results
+        ).invoke(query)
     return Command(
         update={
             "background_investigation_results": json.dumps(
@@ -403,7 +407,6 @@ async def _setup_and_execute_agent_step(
     state: State,
     config: RunnableConfig,
     agent_type: str,
-    default_agent,
     default_tools: list,
 ) -> Command[Literal["research_team"]]:
     """Helper function to set up an agent with appropriate tools and execute a step.
@@ -417,7 +420,6 @@ async def _setup_and_execute_agent_step(
         state: The current state
         config: The runnable config
         agent_type: The type of agent ("researcher" or "coder")
-        default_agent: The default agent to use if no MCP servers are configured
         default_tools: The default tools to add to the agent
 
     Returns:
@@ -455,8 +457,9 @@ async def _setup_and_execute_agent_step(
             agent = create_agent(agent_type, agent_type, loaded_tools, agent_type)
             return await _execute_agent_step(state, agent, agent_type)
     else:
-        # Use default agent if no MCP servers are configured
-        return await _execute_agent_step(state, default_agent, agent_type)
+        # Use default tools if no MCP servers are configured
+        agent = create_agent(agent_type, agent_type, default_tools, agent_type)
+        return await _execute_agent_step(state, agent, agent_type)
 
 
 async def researcher_node(
@@ -464,12 +467,12 @@ async def researcher_node(
 ) -> Command[Literal["research_team"]]:
     """Researcher node that do research"""
     logger.info("Researcher node is researching.")
+    configurable = Configuration.from_runnable_config(config)
     return await _setup_and_execute_agent_step(
         state,
         config,
         "researcher",
-        research_agent,
-        [web_search_tool, crawl_tool],
+        [get_web_search_tool(configurable.max_search_results), crawl_tool],
     )
 
 
@@ -482,6 +485,5 @@ async def coder_node(
         state,
         config,
         "coder",
-        coder_agent,
         [python_repl_tool],
     )

src/server/app.py

@@ -61,6 +61,7 @@ async def chat_stream(request: ChatRequest):
             thread_id,
             request.max_plan_iterations,
             request.max_step_num,
+            request.max_search_results,
             request.auto_accepted_plan,
             request.interrupt_feedback,
             request.mcp_settings,
@@ -75,6 +76,7 @@ async def _astream_workflow_generator(
     thread_id: str,
     max_plan_iterations: int,
     max_step_num: int,
+    max_search_results: int,
     auto_accepted_plan: bool,
     interrupt_feedback: str,
     mcp_settings: dict,
@@ -101,6 +103,7 @@ async def _astream_workflow_generator(
             "thread_id": thread_id,
             "max_plan_iterations": max_plan_iterations,
             "max_step_num": max_step_num,
+            "max_search_results": max_search_results,
             "mcp_settings": mcp_settings,
         },
         stream_mode=["messages", "updates"],

src/server/chat_request.py

@@ -38,6 +38,9 @@ class ChatRequest(BaseModel):
     max_step_num: Optional[int] = Field(
         3, description="The maximum number of steps in a plan"
     )
+    max_search_results: Optional[int] = Field(
+        3, description="The maximum number of search results"
+    )
     auto_accepted_plan: Optional[bool] = Field(
         False, description="Whether to automatically accept the plan"
     )

src/tools/__init__.py

@@ -5,28 +5,12 @@ import os
 
 from .crawl import crawl_tool
 from .python_repl import python_repl_tool
-from .search import (
-    tavily_search_tool,
-    duckduckgo_search_tool,
-    brave_search_tool,
-    arxiv_search_tool,
-)
+from .search import get_web_search_tool
 from .tts import VolcengineTTS
-from src.config import SELECTED_SEARCH_ENGINE, SearchEngine
-
-# Map search engine names to their respective tools
-search_tool_mappings = {
-    SearchEngine.TAVILY.value: tavily_search_tool,
-    SearchEngine.DUCKDUCKGO.value: duckduckgo_search_tool,
-    SearchEngine.BRAVE_SEARCH.value: brave_search_tool,
-    SearchEngine.ARXIV.value: arxiv_search_tool,
-}
-
-web_search_tool = search_tool_mappings.get(SELECTED_SEARCH_ENGINE, tavily_search_tool)
 
 __all__ = [
     "crawl_tool",
-    "web_search_tool",
     "python_repl_tool",
+    "get_web_search_tool",
     "VolcengineTTS",
 ]

src/tools/search.py

@@ -9,7 +9,7 @@ from langchain_community.tools import BraveSearch, DuckDuckGoSearchResults
 from langchain_community.tools.arxiv import ArxivQueryRun
 from langchain_community.utilities import ArxivAPIWrapper, BraveSearchWrapper
 
-from src.config import SEARCH_MAX_RESULTS, SearchEngine
+from src.config import SearchEngine, SELECTED_SEARCH_ENGINE
 from src.tools.tavily_search.tavily_search_results_with_images import (
     TavilySearchResultsWithImages,
 )
@@ -18,44 +18,48 @@ from src.tools.decorators import create_logged_tool
 
 logger = logging.getLogger(__name__)
 
+# Create logged versions of the search tools
 LoggedTavilySearch = create_logged_tool(TavilySearchResultsWithImages)
-if os.getenv("SEARCH_API", "") == SearchEngine.TAVILY.value:
-    tavily_search_tool = LoggedTavilySearch(
-        name="web_search",
-        max_results=SEARCH_MAX_RESULTS,
-        include_raw_content=True,
-        include_images=True,
-        include_image_descriptions=True,
-    )
-else:
-    tavily_search_tool = None
-
 LoggedDuckDuckGoSearch = create_logged_tool(DuckDuckGoSearchResults)
-duckduckgo_search_tool = LoggedDuckDuckGoSearch(
-    name="web_search", max_results=SEARCH_MAX_RESULTS
-)
-
 LoggedBraveSearch = create_logged_tool(BraveSearch)
-brave_search_tool = LoggedBraveSearch(
-    name="web_search",
-    search_wrapper=BraveSearchWrapper(
-        api_key=os.getenv("BRAVE_SEARCH_API_KEY", ""),
-        search_kwargs={"count": SEARCH_MAX_RESULTS},
-    ),
-)
-
 LoggedArxivSearch = create_logged_tool(ArxivQueryRun)
-arxiv_search_tool = LoggedArxivSearch(
-    name="web_search",
-    api_wrapper=ArxivAPIWrapper(
-        top_k_results=SEARCH_MAX_RESULTS,
-        load_max_docs=SEARCH_MAX_RESULTS,
-        load_all_available_meta=True,
-    ),
-)
+
+
+# Get the selected search tool
+def get_web_search_tool(max_search_results: int):
+    if SELECTED_SEARCH_ENGINE == SearchEngine.TAVILY.value:
+        return LoggedTavilySearch(
+            name="web_search",
+            max_results=max_search_results,
+            include_raw_content=True,
+            include_images=True,
+            include_image_descriptions=True,
+        )
+    elif SELECTED_SEARCH_ENGINE == SearchEngine.DUCKDUCKGO.value:
+        return LoggedDuckDuckGoSearch(name="web_search", max_results=max_search_results)
+    elif SELECTED_SEARCH_ENGINE == SearchEngine.BRAVE_SEARCH.value:
+        return LoggedBraveSearch(
+            name="web_search",
+            search_wrapper=BraveSearchWrapper(
+                api_key=os.getenv("BRAVE_SEARCH_API_KEY", ""),
+                search_kwargs={"count": max_search_results},
+            ),
+        )
+    elif SELECTED_SEARCH_ENGINE == SearchEngine.ARXIV.value:
+        return LoggedArxivSearch(
+            name="web_search",
+            api_wrapper=ArxivAPIWrapper(
+                top_k_results=max_search_results,
+                load_max_docs=max_search_results,
+                load_all_available_meta=True,
+            ),
+        )
+    else:
+        raise ValueError(f"Unsupported search engine: {SELECTED_SEARCH_ENGINE}")
+
 
 if __name__ == "__main__":
     results = LoggedDuckDuckGoSearch(
-        name="web_search", max_results=SEARCH_MAX_RESULTS, output_format="list"
+        name="web_search", max_results=3, output_format="list"
     ).invoke("cute panda")
     print(json.dumps(results, indent=2, ensure_ascii=False))

web/src/app/settings/tabs/general-tab.tsx

@@ -32,6 +32,9 @@ const generalFormSchema = z.object({
   maxStepNum: z.number().min(1, {
     message: "Max step number must be at least 1.",
   }),
+  maxSearchResults: z.number().min(1, {
+    message: "Max search results must be at least 1.",
+  }),
 });
 
 export const GeneralTab: Tab = ({
@@ -143,6 +146,30 @@ export const GeneralTab: Tab = ({
                 </FormItem>
               )}
             />
+            <FormField
+              control={form.control}
+              name="maxSearchResults"
+              render={({ field }) => (
+                <FormItem>
+                  <FormLabel>Max search results</FormLabel>
+                  <FormControl>
+                    <Input
+                      className="w-60"
+                      type="number"
+                      defaultValue={field.value}
+                      min={1}
+                      onChange={(event) =>
+                        field.onChange(parseInt(event.target.value || "0"))
+                      }
+                    />
+                  </FormControl>
+                  <FormDescription>
+                    By default, each search step has 3 results.
+                  </FormDescription>
+                  <FormMessage />
+                </FormItem>
+              )}
+            />
           </form>
         </Form>
       </main>

web/src/core/api/chat.ts

@@ -18,6 +18,7 @@ export async function* chatStream(
     auto_accepted_plan: boolean;
     max_plan_iterations: number;
     max_step_num: number;
+    max_search_results?: number;
     interrupt_feedback?: string;
     enable_background_investigation: boolean;
     mcp_settings?: {
@@ -61,12 +62,14 @@ async function* chatReplayStream(
     auto_accepted_plan: boolean;
     max_plan_iterations: number;
     max_step_num: number;
+    max_search_results?: number;
     interrupt_feedback?: string;
   } = {
     thread_id: "__mock__",
     auto_accepted_plan: false,
     max_plan_iterations: 3,
     max_step_num: 1,
+    max_search_results: 3,
     interrupt_feedback: undefined,
   },
   options: { abortSignal?: AbortSignal } = {},
@@ -157,6 +160,7 @@ export async function fetchReplayTitle() {
       auto_accepted_plan: false,
       max_plan_iterations: 3,
       max_step_num: 1,
+      max_search_results: 3,
     },
     {},
   );

web/src/core/store/settings-store.ts

@@ -13,6 +13,7 @@ const DEFAULT_SETTINGS: SettingsState = {
     enableBackgroundInvestigation: false,
     maxPlanIterations: 1,
     maxStepNum: 3,
+    maxSearchResults: 3,
   },
   mcp: {
     servers: [],
@@ -25,6 +26,7 @@ export type SettingsState = {
     enableBackgroundInvestigation: boolean;
     maxPlanIterations: number;
     maxStepNum: number;
+    maxSearchResults: number;
   };
   mcp: {
     servers: MCPServerMetadata[];

web/src/core/store/store.ts

@@ -104,6 +104,7 @@ export async function sendMessage(
         settings.enableBackgroundInvestigation ?? true,
       max_plan_iterations: settings.maxPlanIterations,
       max_step_num: settings.maxStepNum,
+      max_search_results: settings.maxSearchResults,
       mcp_settings: settings.mcpSettings,
     },
     options,