Commit ·
a394be7
1
Parent(s): 0019780
refactor(agent): split websearch agent into separate modules
Browse filesSplit monolithic websearch.py into websearchagent.py and websearchagents.py
to improve code organization and maintainability. The supervisor agent now
has access to both individual websearch_agent and aggregated web_search_agents.
- agent/agent.py +4 -2
- agent/agents/__init__.py +3 -2
- agent/agents/{websearch.py → websearchagent.py} +19 -5
- agent/agents/websearchagents.py +309 -0
- agent/tools/search.py +1 -1
agent/agent.py
CHANGED
|
@@ -4,7 +4,9 @@ from colorama import Fore, Style # type: ignore[import]
|
|
| 4 |
from langchain.agents import create_agent
|
| 5 |
from langchain_core.messages import HumanMessage
|
| 6 |
from agent.tools.math_solver import math_solver
|
| 7 |
-
|
|
|
|
|
|
|
| 8 |
|
| 9 |
load_dotenv()
|
| 10 |
|
|
@@ -13,7 +15,7 @@ def supervisor_agent():
|
|
| 13 |
"""Return a supervisor agent instance with math_solver and websearch_agent."""
|
| 14 |
return create_agent(
|
| 15 |
model="google_genai:gemini-3-flash-preview",
|
| 16 |
-
tools=[math_solver, websearch_agent],
|
| 17 |
system_prompt=(
|
| 18 |
f"You are a supervisor agent. "
|
| 19 |
f"Current time is: {datetime.now(timezone.utc).isoformat()}. "
|
|
|
|
| 4 |
from langchain.agents import create_agent
|
| 5 |
from langchain_core.messages import HumanMessage
|
| 6 |
from agent.tools.math_solver import math_solver
|
| 7 |
+
|
| 8 |
+
from agent.agents.websearchagents import web_search_agents
|
| 9 |
+
from agent.agents.websearchagent import websearch_agent
|
| 10 |
|
| 11 |
load_dotenv()
|
| 12 |
|
|
|
|
| 15 |
"""Return a supervisor agent instance with math_solver and websearch_agent."""
|
| 16 |
return create_agent(
|
| 17 |
model="google_genai:gemini-3-flash-preview",
|
| 18 |
+
tools=[math_solver, websearch_agent, web_search_agents],
|
| 19 |
system_prompt=(
|
| 20 |
f"You are a supervisor agent. "
|
| 21 |
f"Current time is: {datetime.now(timezone.utc).isoformat()}. "
|
agent/agents/__init__.py
CHANGED
|
@@ -1,3 +1,4 @@
|
|
| 1 |
-
from agent.agents.
|
|
|
|
| 2 |
|
| 3 |
-
__all__ = ["websearch_agent"]
|
|
|
|
| 1 |
+
from agent.agents.websearchagent import websearch_agent
|
| 2 |
+
from agent.agents.websearchagents import web_search_agents
|
| 3 |
|
| 4 |
+
__all__ = ["websearch_agent", "web_search_agents"]
|
agent/agents/{websearch.py → websearchagent.py}
RENAMED
|
@@ -9,10 +9,17 @@ from agent.tools.search import web_search
|
|
| 9 |
@tool
|
| 10 |
def websearch_agent(query: str) -> str:
|
| 11 |
"""
|
| 12 |
-
A web search agent that searches the internet and returns an answer.
|
| 13 |
-
Use this tool when you need to find real-time or factual information
|
| 14 |
-
|
| 15 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
|
| 17 |
Args:
|
| 18 |
query: The question or search query to look up on the web.
|
|
@@ -32,7 +39,7 @@ def websearch_agent(query: str) -> str:
|
|
| 32 |
try:
|
| 33 |
result = base_agent.invoke(
|
| 34 |
{"messages": [{"role": "user", "content": query}]},
|
| 35 |
-
config={"recursion_limit":
|
| 36 |
)
|
| 37 |
content = result["messages"][-1].content
|
| 38 |
if isinstance(content, list):
|
|
@@ -44,6 +51,13 @@ def websearch_agent(query: str) -> str:
|
|
| 44 |
f"{Fore.RED}[WebSearchAgent] Recursion limit reached, returning partial results.{Style.RESET_ALL}"
|
| 45 |
)
|
| 46 |
content = "Search completed but no definitive answer was found within the allowed steps."
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 47 |
print(
|
| 48 |
f"{Fore.YELLOW}[WebSearchAgent -> SupervisorAgent] {content}{Style.RESET_ALL}"
|
| 49 |
)
|
|
|
|
| 9 |
@tool
|
| 10 |
def websearch_agent(query: str) -> str:
|
| 11 |
"""
|
| 12 |
+
A single web search agent that searches the internet and returns an answer.
|
| 13 |
+
Use this tool when you need to find real-time or factual information from the web.
|
| 14 |
+
|
| 15 |
+
Pros:
|
| 16 |
+
- Has continuous memory across search steps, allowing deep investigation on a single topic.
|
| 17 |
+
Cons:
|
| 18 |
+
- Narrow field of view, can only follow one search thread at a time.
|
| 19 |
+
- May fail after too many steps due to token limit overflow.
|
| 20 |
+
|
| 21 |
+
Prefer websearch_agents for complex questions requiring broad, multi-source research.
|
| 22 |
+
Use this tool for simple, direct factual lookups.
|
| 23 |
|
| 24 |
Args:
|
| 25 |
query: The question or search query to look up on the web.
|
|
|
|
| 39 |
try:
|
| 40 |
result = base_agent.invoke(
|
| 41 |
{"messages": [{"role": "user", "content": query}]},
|
| 42 |
+
# config={"recursion_limit": 10},
|
| 43 |
)
|
| 44 |
content = result["messages"][-1].content
|
| 45 |
if isinstance(content, list):
|
|
|
|
| 51 |
f"{Fore.RED}[WebSearchAgent] Recursion limit reached, returning partial results.{Style.RESET_ALL}"
|
| 52 |
)
|
| 53 |
content = "Search completed but no definitive answer was found within the allowed steps."
|
| 54 |
+
except Exception as e:
|
| 55 |
+
error_msg = str(e)
|
| 56 |
+
print(f"{Fore.RED}[WebSearchAgent] Error: {error_msg}{Style.RESET_ALL}")
|
| 57 |
+
content = (
|
| 58 |
+
f"Search agent failed with error: {error_msg}. "
|
| 59 |
+
f"Recommend retrying with the web_search_agents tool to avoid context length overflow."
|
| 60 |
+
)
|
| 61 |
print(
|
| 62 |
f"{Fore.YELLOW}[WebSearchAgent -> SupervisorAgent] {content}{Style.RESET_ALL}"
|
| 63 |
)
|
agent/agents/websearchagents.py
ADDED
|
@@ -0,0 +1,309 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import random
|
| 3 |
+
from concurrent.futures import ThreadPoolExecutor, as_completed
|
| 4 |
+
from datetime import datetime, timezone
|
| 5 |
+
from colorama import Fore, Style # type: ignore[import]
|
| 6 |
+
from langchain.agents import create_agent
|
| 7 |
+
from langchain_core.tools import tool
|
| 8 |
+
from pydantic import BaseModel, Field
|
| 9 |
+
from tavily import TavilyClient # type: ignore[import]
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
SUBAGENT_COLORS = [
|
| 13 |
+
Fore.MAGENTA,
|
| 14 |
+
Fore.CYAN,
|
| 15 |
+
Fore.GREEN,
|
| 16 |
+
Fore.YELLOW,
|
| 17 |
+
Fore.BLUE,
|
| 18 |
+
Fore.WHITE,
|
| 19 |
+
Fore.LIGHTRED_EX,
|
| 20 |
+
Fore.LIGHTGREEN_EX,
|
| 21 |
+
Fore.LIGHTYELLOW_EX,
|
| 22 |
+
Fore.LIGHTBLUE_EX,
|
| 23 |
+
Fore.LIGHTMAGENTA_EX,
|
| 24 |
+
Fore.LIGHTCYAN_EX,
|
| 25 |
+
]
|
| 26 |
+
|
| 27 |
+
MAX_CHARS = 900000
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
class ExpandedQueries(BaseModel):
|
| 31 |
+
"""A list of expanded search queries derived from the original query."""
|
| 32 |
+
|
| 33 |
+
queries: list[str] = Field(
|
| 34 |
+
description="A list of expanded search queries to cover different angles of the original question."
|
| 35 |
+
)
|
| 36 |
+
|
| 37 |
+
|
| 38 |
+
# ---------------------------------------------------------------------------
|
| 39 |
+
# Step 1: Query Expansion (structured output)
|
| 40 |
+
# ---------------------------------------------------------------------------
|
| 41 |
+
|
| 42 |
+
|
| 43 |
+
def expand_queries(origin_question: str, query: str) -> list[str]:
|
| 44 |
+
"""Use structured output to expand a single query into multiple search queries."""
|
| 45 |
+
print(f"{Fore.CYAN}[QueryExpander] Expanding: {query}{Style.RESET_ALL}")
|
| 46 |
+
|
| 47 |
+
agent = create_agent(
|
| 48 |
+
model="google_genai:gemini-3-flash-preview",
|
| 49 |
+
response_format=ExpandedQueries,
|
| 50 |
+
system_prompt=(
|
| 51 |
+
f"Current time is: {datetime.now(timezone.utc).isoformat()}. "
|
| 52 |
+
"You are a search query expansion expert. "
|
| 53 |
+
"Given a user question, generate 3 diverse and specific search queries "
|
| 54 |
+
"that cover different angles of the question to maximize search coverage. "
|
| 55 |
+
"Each query should be concise and optimized for web search engines."
|
| 56 |
+
),
|
| 57 |
+
)
|
| 58 |
+
|
| 59 |
+
result = agent.invoke(
|
| 60 |
+
{
|
| 61 |
+
"messages": [
|
| 62 |
+
{
|
| 63 |
+
"role": "user",
|
| 64 |
+
"content": (
|
| 65 |
+
f"Original question: {origin_question}\n"
|
| 66 |
+
f"Query to expand: {query}"
|
| 67 |
+
),
|
| 68 |
+
}
|
| 69 |
+
]
|
| 70 |
+
}
|
| 71 |
+
)
|
| 72 |
+
expanded: ExpandedQueries = result["structured_response"]
|
| 73 |
+
|
| 74 |
+
for i, q in enumerate(expanded.queries, 1):
|
| 75 |
+
print(f"{Fore.CYAN} [{i}] {q}{Style.RESET_ALL}")
|
| 76 |
+
|
| 77 |
+
return expanded.queries
|
| 78 |
+
|
| 79 |
+
|
| 80 |
+
# ---------------------------------------------------------------------------
|
| 81 |
+
# Step 2: Parallel Tavily Search & Extract
|
| 82 |
+
# ---------------------------------------------------------------------------
|
| 83 |
+
|
| 84 |
+
|
| 85 |
+
def _search_single_query(query: str) -> list[dict]:
|
| 86 |
+
"""Search a single query via Tavily and return results with full content."""
|
| 87 |
+
client = TavilyClient(api_key=os.environ["TAVILY_API_KEY"])
|
| 88 |
+
|
| 89 |
+
search_response = client.search(query=query, search_depth="advanced", max_results=3)
|
| 90 |
+
results = search_response.get("results", [])
|
| 91 |
+
if not results:
|
| 92 |
+
return []
|
| 93 |
+
|
| 94 |
+
urls = [r["url"] for r in results]
|
| 95 |
+
try:
|
| 96 |
+
extraction = client.extract(
|
| 97 |
+
urls=urls, extract_depth="advanced", format="markdown"
|
| 98 |
+
)
|
| 99 |
+
extracted_map = {
|
| 100 |
+
item["url"]: item["raw_content"] for item in extraction.get("results", [])
|
| 101 |
+
}
|
| 102 |
+
except Exception as e:
|
| 103 |
+
print(f"{Fore.RED} Extraction failed: {e}{Style.RESET_ALL}")
|
| 104 |
+
extracted_map = {}
|
| 105 |
+
|
| 106 |
+
return [
|
| 107 |
+
{
|
| 108 |
+
"url": r["url"],
|
| 109 |
+
"title": r["title"],
|
| 110 |
+
"snippet": r["content"],
|
| 111 |
+
"full_content": extracted_map.get(r["url"], "Extraction failed."),
|
| 112 |
+
}
|
| 113 |
+
for r in results
|
| 114 |
+
]
|
| 115 |
+
|
| 116 |
+
|
| 117 |
+
def search_and_extract_parallel(queries: list[str]) -> list[dict]:
|
| 118 |
+
"""Search all expanded queries in parallel threads and deduplicate by URL."""
|
| 119 |
+
seen_urls: set[str] = set()
|
| 120 |
+
all_results: list[dict] = []
|
| 121 |
+
|
| 122 |
+
with ThreadPoolExecutor(max_workers=len(queries)) as pool:
|
| 123 |
+
futures = {pool.submit(_search_single_query, q): q for q in queries}
|
| 124 |
+
for future in as_completed(futures):
|
| 125 |
+
q = futures[future]
|
| 126 |
+
print(f"{Fore.GREEN}[Search & Extract] Done: {q}{Style.RESET_ALL}")
|
| 127 |
+
try:
|
| 128 |
+
for item in future.result():
|
| 129 |
+
if item["url"] not in seen_urls:
|
| 130 |
+
seen_urls.add(item["url"])
|
| 131 |
+
all_results.append(item)
|
| 132 |
+
except Exception as e:
|
| 133 |
+
print(f"{Fore.RED}[Search & Extract] Error: {e}{Style.RESET_ALL}")
|
| 134 |
+
|
| 135 |
+
print(
|
| 136 |
+
f"{Fore.GREEN}[SearchAgents] Collected {len(all_results)} unique pages.{Style.RESET_ALL}"
|
| 137 |
+
)
|
| 138 |
+
return all_results
|
| 139 |
+
|
| 140 |
+
|
| 141 |
+
# ---------------------------------------------------------------------------
|
| 142 |
+
# Step 3: SubAgent — investigate a single page
|
| 143 |
+
# ---------------------------------------------------------------------------
|
| 144 |
+
|
| 145 |
+
|
| 146 |
+
def subagent(origin_question: str, query: str) -> str:
|
| 147 |
+
"""
|
| 148 |
+
Investigate a single URL's full content against the original question.
|
| 149 |
+
The query parameter contains the URL + full page content (truncated to MAX_CHARS).
|
| 150 |
+
Returns the agent's findings as a string.
|
| 151 |
+
"""
|
| 152 |
+
prompt = (f"Original question: {origin_question}\n\nWeb page content:\n{query}")[
|
| 153 |
+
:MAX_CHARS
|
| 154 |
+
]
|
| 155 |
+
|
| 156 |
+
color = random.choice(SUBAGENT_COLORS)
|
| 157 |
+
print(f"{color}[SubAgent] Investigating ({len(prompt)} chars)...{Style.RESET_ALL}")
|
| 158 |
+
|
| 159 |
+
agent = create_agent(
|
| 160 |
+
model="google_genai:gemini-3-flash-preview",
|
| 161 |
+
system_prompt=(
|
| 162 |
+
f"Current time is: {datetime.now(timezone.utc).isoformat()}. "
|
| 163 |
+
"You are a research analyst. You are given a web page's full content "
|
| 164 |
+
"and an original question. Extract ALL relevant clues, facts, data, "
|
| 165 |
+
"and details from the page that help answer the original question. "
|
| 166 |
+
"Be thorough and precise. Include specific numbers, names, and dates."
|
| 167 |
+
),
|
| 168 |
+
)
|
| 169 |
+
|
| 170 |
+
result = agent.invoke({"messages": [{"role": "user", "content": prompt}]})
|
| 171 |
+
content = result["messages"][-1].content
|
| 172 |
+
if isinstance(content, list):
|
| 173 |
+
content = content[0].get("text", "")
|
| 174 |
+
return str(content)
|
| 175 |
+
|
| 176 |
+
|
| 177 |
+
# ---------------------------------------------------------------------------
|
| 178 |
+
# Step 4: Combine all subagent findings
|
| 179 |
+
# ---------------------------------------------------------------------------
|
| 180 |
+
|
| 181 |
+
|
| 182 |
+
def combine_result_agent(origin_question: str, query: str) -> str:
|
| 183 |
+
"""
|
| 184 |
+
Combine multiple subagent findings into a single comprehensive answer.
|
| 185 |
+
The query parameter contains all subagent outputs joined together (truncated to MAX_CHARS).
|
| 186 |
+
"""
|
| 187 |
+
prompt = (
|
| 188 |
+
f"Original question: {origin_question}\n\n"
|
| 189 |
+
f"Research findings from multiple sources:\n{query}"
|
| 190 |
+
)[:MAX_CHARS]
|
| 191 |
+
|
| 192 |
+
print(
|
| 193 |
+
f"{Fore.BLUE}[CombineAgent] Synthesizing ({len(prompt)} chars)...{Style.RESET_ALL}"
|
| 194 |
+
)
|
| 195 |
+
|
| 196 |
+
agent = create_agent(
|
| 197 |
+
model="google_genai:gemini-3-flash-preview",
|
| 198 |
+
system_prompt=(
|
| 199 |
+
f"Current time is: {datetime.now(timezone.utc).isoformat()}. "
|
| 200 |
+
"You are a research synthesizer. You receive findings from multiple "
|
| 201 |
+
"web sources investigating a question. Combine them into a single, "
|
| 202 |
+
"comprehensive, well-structured answer. Cite the source URL for each "
|
| 203 |
+
"key fact. Resolve any contradictions between sources."
|
| 204 |
+
),
|
| 205 |
+
)
|
| 206 |
+
|
| 207 |
+
result = agent.invoke({"messages": [{"role": "user", "content": prompt}]})
|
| 208 |
+
content = result["messages"][-1].content
|
| 209 |
+
if isinstance(content, list):
|
| 210 |
+
content = content[0].get("text", "")
|
| 211 |
+
return str(content)
|
| 212 |
+
|
| 213 |
+
|
| 214 |
+
# ---------------------------------------------------------------------------
|
| 215 |
+
# Step 5: Main orchestrator tool
|
| 216 |
+
# ---------------------------------------------------------------------------
|
| 217 |
+
|
| 218 |
+
|
| 219 |
+
@tool
|
| 220 |
+
def web_search_agents(origin_question: str, query: str) -> str:
|
| 221 |
+
"""
|
| 222 |
+
A multi-agent web search tool that expands the query, searches in parallel,
|
| 223 |
+
investigates each page with subagents, and synthesizes a final answer.
|
| 224 |
+
|
| 225 |
+
Pros:
|
| 226 |
+
- Dispatches multiple subagents for deep, parallel investigation across many sources.
|
| 227 |
+
- Can achieve both broad and deep research when queries are well-crafted.
|
| 228 |
+
Cons:
|
| 229 |
+
- Requires more detailed and transparent query descriptions for good control.
|
| 230 |
+
- Each subagent has no long-term memory (context is kept short to avoid token limit failures).
|
| 231 |
+
|
| 232 |
+
Use this tool for complex questions that require deep web research from multiple sources.
|
| 233 |
+
For simple factual lookups, prefer websearch_agent instead.
|
| 234 |
+
|
| 235 |
+
Args:
|
| 236 |
+
origin_question: The original user question for context. Must be detailed and clear.
|
| 237 |
+
query: The specific search query to research. Be as specific and transparent as possible.
|
| 238 |
+
"""
|
| 239 |
+
print(f"\n{Fore.YELLOW}{'=' * 60}")
|
| 240 |
+
print("[WebSearchAgents] Starting research")
|
| 241 |
+
print(f" Origin : {origin_question}")
|
| 242 |
+
print(f" Query : {query}")
|
| 243 |
+
print(f"{'=' * 60}{Style.RESET_ALL}\n")
|
| 244 |
+
|
| 245 |
+
# 1. Expand queries
|
| 246 |
+
expanded = expand_queries(origin_question, query)
|
| 247 |
+
|
| 248 |
+
# 2. Parallel Tavily search & extract
|
| 249 |
+
pages = search_and_extract_parallel(expanded)
|
| 250 |
+
if not pages:
|
| 251 |
+
return "No search results found."
|
| 252 |
+
|
| 253 |
+
# 3. Parallel subagent investigation
|
| 254 |
+
print(
|
| 255 |
+
f"\n{Fore.MAGENTA}[SubAgents] Dispatching {len(pages)} subagents...{Style.RESET_ALL}"
|
| 256 |
+
)
|
| 257 |
+
subagent_results: list[str] = []
|
| 258 |
+
|
| 259 |
+
def _run_subagent(page: dict) -> str:
|
| 260 |
+
page_input = (
|
| 261 |
+
f"URL: {page['url']}\nTitle: {page['title']}\n\n{page['full_content']}"
|
| 262 |
+
)
|
| 263 |
+
finding = subagent(origin_question, page_input)
|
| 264 |
+
return f"### Source: {page['url']}\n{finding}"
|
| 265 |
+
|
| 266 |
+
with ThreadPoolExecutor(max_workers=min(len(pages), 5)) as pool:
|
| 267 |
+
futures = {pool.submit(_run_subagent, p): p for p in pages}
|
| 268 |
+
for future in as_completed(futures):
|
| 269 |
+
page = futures[future]
|
| 270 |
+
try:
|
| 271 |
+
result = future.result()
|
| 272 |
+
subagent_results.append(result)
|
| 273 |
+
color = random.choice(SUBAGENT_COLORS)
|
| 274 |
+
print(f"{color}[SubAgent] Done: {page['url']}{Style.RESET_ALL}")
|
| 275 |
+
except Exception as e:
|
| 276 |
+
print(
|
| 277 |
+
f"{Fore.RED}[SubAgent] Error on {page['url']}: {e}{Style.RESET_ALL}"
|
| 278 |
+
)
|
| 279 |
+
|
| 280 |
+
# 4. Combine results
|
| 281 |
+
combined_input = "\n\n---\n\n".join(subagent_results)
|
| 282 |
+
result = combine_result_agent(origin_question, combined_input)
|
| 283 |
+
|
| 284 |
+
print(f"\n{Fore.YELLOW}{'=' * 60}")
|
| 285 |
+
print("[WebSearchAgents] Research complete")
|
| 286 |
+
print(f"{'=' * 60}")
|
| 287 |
+
print(f"[WebSearchAgents -> SupervisorAgent] {result}")
|
| 288 |
+
print(f"{'=' * 60}{Style.RESET_ALL}\n")
|
| 289 |
+
|
| 290 |
+
return result
|
| 291 |
+
|
| 292 |
+
|
| 293 |
+
# ---------------------------------------------------------------------------
|
| 294 |
+
# Test
|
| 295 |
+
# ---------------------------------------------------------------------------
|
| 296 |
+
|
| 297 |
+
if __name__ == "__main__":
|
| 298 |
+
from dotenv import load_dotenv
|
| 299 |
+
|
| 300 |
+
load_dotenv()
|
| 301 |
+
|
| 302 |
+
test_query = "What is LangGraph?"
|
| 303 |
+
answer = web_search_agents.invoke(
|
| 304 |
+
{"origin_question": test_query, "query": test_query}
|
| 305 |
+
)
|
| 306 |
+
print(f"\n{Fore.YELLOW}{'=' * 60}")
|
| 307 |
+
print("FINAL ANSWER")
|
| 308 |
+
print(f"{'=' * 60}{Style.RESET_ALL}")
|
| 309 |
+
print(answer)
|
agent/tools/search.py
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
import os
|
| 2 |
-
from colorama import Fore, Style
|
| 3 |
from langchain_core.tools import tool
|
| 4 |
from tavily import TavilyClient # type: ignore[import]
|
| 5 |
|
|
|
|
| 1 |
import os
|
| 2 |
+
from colorama import Fore, Style # type: ignore[import]
|
| 3 |
from langchain_core.tools import tool
|
| 4 |
from tavily import TavilyClient # type: ignore[import]
|
| 5 |
|