Spaces:
Sleeping
Sleeping
File size: 18,566 Bytes
d6ea378 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 | """Deep Research Multi-Agent System implementation.
This implementation focuses on:
- Building a multi-agent system for comprehensive research
- Using LangGraph with MessagesState for proper state management
- Synthesizing research findings into structured reports
"""
from typing import Dict, List, Optional, Any, Annotated
from langchain_core.messages import HumanMessage, AIMessage, BaseMessage
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_openai import ChatOpenAI
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_core.tools import tool
from langgraph.graph import StateGraph, END, START, MessagesState
from langgraph.prebuilt import create_react_agent
from pydantic import BaseModel, Field
from core.chat_interface import ChatInterface
from opik.integrations.langchain import OpikTracer
from agents.prompts import (
RESEARCH_MANAGER_PROMPT,
REPORT_FINALIZER_PROMPT,
)
from dotenv import load_dotenv
load_dotenv()
class ResearchQuestion(BaseModel):
"""A research question with a title and description."""
title: str = Field(description="The title of the research question/section")
description: str = Field(description="Description of what to research for this section")
completed: bool = Field(default=False, description="Whether research has been completed for this section")
class ResearchPlan(BaseModel):
"""The overall research plan created by the Research Manager."""
topic: str = Field(description="The main research topic")
questions: List[ResearchQuestion] = Field(description="The list of research questions to investigate")
current_question_index: int = Field(default=0, description="Index of the current question being researched")
class Report(BaseModel):
"""The final research report structure."""
executive_summary: Optional[str] = Field(default=None, description="Executive summary of the research")
key_findings: Optional[str] = Field(default=None, description="Key findings from the research")
detailed_analysis: List[Dict[str, Any]] = Field(default_factory=list, description="Detailed analysis sections")
limitations: Optional[str] = Field(default=None, description="Limitations and further research")
class ResearchState(MessagesState):
"""State tracking for the deep research workflow using MessagesState as base."""
research_plan: Optional[ResearchPlan] = None
report: Optional[Report] = None
next_step: str = "research_manager"
# MessagesState already handles messages with add_messages reducer
class DeepResearchChat(ChatInterface):
"""Deep research implementation using multi-agent system with proper state management."""
def __init__(self):
self.llm = None
self.research_manager = None
self.specialized_research_agent = None
self.finalizer = None
self.workflow = None
self.tavily_search_tool = None
def initialize(self) -> None:
"""Initialize components for the deep research system."""
# Initialize LLM model
self.llm = ChatOpenAI(model="gpt-4o", temperature=0)
# Create Tavily search tool for agents
self.tavily_search_tool = TavilySearchResults(max_results=5)
# Create components
self.research_manager = self._create_research_manager()
self.specialized_research_agent = self._create_specialized_research_agent()
self.finalizer = self._create_finalizer()
# Create the workflow graph using these agents
self.workflow = self._create_workflow()
# Optional: Create Opik Tracer for monitoring
try:
self.tracer = OpikTracer(
graph=self.workflow.get_graph(xray=True),
project_name="nexus-research-workflow"
)
except:
self.tracer = None
print("Opik tracer not available, continuing without monitoring")
def _create_research_manager(self) -> Any:
"""Create the research manager agent."""
research_manager = (
RESEARCH_MANAGER_PROMPT
| self.llm.with_structured_output(ResearchPlan)
)
return research_manager
def _create_specialized_research_agent(self) -> Any:
"""Create specialized research agents."""
# Create search tool for the agent
@tool("web_search")
def search_web(query: str) -> str:
"""Search the web for information on the research topic."""
results = self.tavily_search_tool.invoke(query)
formatted_results = []
for i, result in enumerate(results, 1):
formatted_results.append(f"Result {i}:")
formatted_results.append(f"Title: {result.get('title', 'N/A')}")
formatted_results.append(f"Content: {result.get('content', 'N/A')}")
formatted_results.append(f"URL: {result.get('url', 'N/A')}")
formatted_results.append("")
return "\n".join(formatted_results)
# Create the specialized agent
tools = [search_web]
# Define the system message for the specialized research agent
system_message = """You are a Specialized Research Agent responsible for thoroughly researching a specific topic section.
Process:
1. Analyze the research question and description
2. Generate effective search queries to gather information
3. Use the web_search tool to find relevant information
4. Synthesize findings into a comprehensive section
5. Include proper citations to your sources
Your response should be:
- Thorough (at least 500 words)
- Well-structured with subsections
- Based on factual information (not made up)
- Include proper citations to sources
Always critically evaluate information and ensure you cover the topic comprehensively.
"""
# Create the specialized research agent
specialized_agent = create_react_agent(
model=self.llm,
tools=tools,
prompt=system_message
)
return specialized_agent
def _create_finalizer(self) -> Any:
"""Create the finalizer component."""
finalizer = REPORT_FINALIZER_PROMPT | self.llm | StrOutputParser()
return finalizer
def _create_workflow(self) -> Any:
"""Create the multi-agent deep research workflow with proper MessagesState usage."""
# Create a state graph
workflow = StateGraph(ResearchState)
# Define the nodes
# Research Manager Node
def research_manager_node(state: ResearchState) -> Dict:
"""Create the research plan and update messages."""
print("\n=== RESEARCH MANAGER NODE ===")
# Get the topic from the LAST user message (not first)
# This handles conversation context better
user_messages = [msg for msg in state["messages"] if isinstance(msg, HumanMessage)]
topic = user_messages[-1].content if user_messages else state["messages"][-1].content
print(f"Planning research for topic: {topic}")
# Generate research plan
research_plan = self.research_manager.invoke({"topic": topic})
print(f"Created research plan with {len(research_plan.questions)} questions")
# Initialize empty report structure
report = Report(
detailed_analysis=[
{"title": q.title, "content": None, "sources": []}
for q in research_plan.questions
]
)
# Add planning message to state
planning_msg = AIMessage(
content=f"Research plan created with {len(research_plan.questions)} sections to investigate."
)
return {
"messages": [planning_msg],
"research_plan": research_plan,
"report": report,
}
# Specialized Research Node
def specialized_research_node(state: ResearchState) -> Dict:
"""Conduct research on the current question and update messages."""
print("\n=== SPECIALIZED RESEARCH NODE ===")
research_plan = state["research_plan"]
assert research_plan is not None, "Research plan is None"
current_index = research_plan.current_question_index
if current_index >= len(research_plan.questions):
print("All research questions completed")
return {}
current_question = research_plan.questions[current_index]
print(f"Researching question {current_index + 1}/{len(research_plan.questions)}: "
f"{current_question.title}")
# Create input for the specialized agent
research_input = {
"messages": [
("user", f"""Research the following topic thoroughly:
Topic: {current_question.title}
Description: {current_question.description}
Provide a detailed analysis with proper citations to sources.
""")
]
}
# Invoke the specialized agent
result = self.specialized_research_agent.invoke(research_input)
# Extract content from the result
last_message = result["messages"][-1]
if isinstance(last_message, tuple):
content = last_message[1]
else:
content = last_message.content
# Parse out sources from the content
sources = []
for line in content.split("\n"):
if "http" in line and "://" in line:
sources.append(line.strip())
# Update the research plan
research_plan.questions[current_index].completed = True
# Update the report
report = state["report"]
assert report is not None, "Report is None"
report.detailed_analysis[current_index]["content"] = content
report.detailed_analysis[current_index]["sources"] = sources
# Move to the next question
research_plan.current_question_index += 1
# Add research progress message
progress_msg = AIMessage(
content=f"Completed research for section: {current_question.title}"
)
return {
"messages": [progress_msg],
"research_plan": research_plan,
"report": report,
}
# Research Evaluator Node
def evaluator_node(state: ResearchState) -> Dict:
"""Evaluate the research progress and determine next steps."""
print("\n=== EVALUATOR NODE ===")
research_plan = state["research_plan"]
assert research_plan is not None, "Research plan is None"
# Check if we've completed all questions
all_completed = research_plan.current_question_index >= len(research_plan.questions)
if all_completed:
print("All research questions have been addressed. Moving to finalizer.")
eval_msg = AIMessage(content="All research sections completed. Finalizing report...")
return {
"messages": [eval_msg],
"next_step": "finalize"
}
else:
# We have more sections to research
next_section = research_plan.questions[research_plan.current_question_index].title
print(f"More research needed. Moving to next section: {next_section}")
eval_msg = AIMessage(content=f"Moving to research section: {next_section}")
return {
"messages": [eval_msg],
"next_step": "research"
}
# Finalizer Node
def finalizer_node(state: ResearchState) -> Dict:
"""Finalize the research report and update messages."""
print("\n=== FINALIZER NODE ===")
research_plan = state["research_plan"]
report = state["report"]
assert report is not None, "Report is None"
assert research_plan is not None, "Research plan is None"
# Prepare the detailed analysis for the finalizer
detailed_analysis = "\n\n".join([
f"## {section['title']}\n{section['content']}"
for section in report.detailed_analysis
if section['content'] is not None
])
# Generate the final sections
final_sections = self.finalizer.invoke({
"topic": research_plan.topic,
"detailed_analysis": detailed_analysis
})
# Parse the final sections
sections = final_sections.split("\n\n")
# Update the report
if len(sections) >= 3:
report.executive_summary = sections[0].replace("# Executive Summary", "").strip()
report.key_findings = sections[1].replace("# Key Findings", "").strip()
report.limitations = sections[2].replace("# Limitations and Further Research", "").strip()
# Format the final report
report_message = self._format_report(report)
return {
"messages": [report_message],
}
# Add nodes to the graph
workflow.add_node("research_manager", research_manager_node)
workflow.add_node("specialized_research", specialized_research_node)
workflow.add_node("evaluate", evaluator_node)
workflow.add_node("finalizer", finalizer_node)
# Add edges
workflow.add_edge(START, "research_manager")
workflow.add_edge("research_manager", "specialized_research")
workflow.add_edge("specialized_research", "evaluate")
# Add conditional edges from evaluator
workflow.add_conditional_edges(
"evaluate",
lambda x: x["next_step"],
{
"research": "specialized_research",
"finalize": "finalizer"
}
)
workflow.add_edge("finalizer", END)
# Compile the workflow
return workflow.compile()
def _format_report(self, report: Report) -> AIMessage:
"""Format the research report for presentation."""
sections = [
"# Research Report\n",
"## Executive Summary\n" + (report.executive_summary or "N/A"),
"## Key Findings\n" + (report.key_findings or "N/A"),
"## Detailed Analysis"
]
# Add detailed analysis sections
for section in report.detailed_analysis:
if section["content"]:
sections.append(f"### {section['title']}\n{section['content']}")
if section["sources"]:
sources = "\n".join([f"- {source}" for source in section["sources"]])
sections.append(f"**Sources:**\n{sources}")
# Add limitations
sections.append("## Limitations and Further Research\n" + (report.limitations or "N/A"))
return AIMessage(content="\n\n".join(sections))
def _convert_history_to_messages(self, chat_history: Optional[List[Dict[str, str]]]) -> List:
"""Convert chat history to LangChain message format.
Args:
chat_history: List of dicts with 'role' and 'content' keys
Returns:
List of LangChain message objects
"""
messages = []
if chat_history:
for msg in chat_history:
if msg["role"] == "user":
messages.append(HumanMessage(content=msg["content"]))
elif msg["role"] == "assistant":
messages.append(AIMessage(content=msg["content"]))
return messages
def process_message(self, message: str, chat_history: Optional[List[Dict[str, str]]] = None) -> str:
"""Process a message using the deep research system with proper state management."""
print("\n=== STARTING DEEP RESEARCH ===")
print(f"Research Topic: {message}")
# Convert chat history to messages
history_messages = self._convert_history_to_messages(chat_history)
# Add the current message
history_messages.append(HumanMessage(content=message))
# Create initial state with full conversation history
initial_state = ResearchState(
messages=history_messages, # Include full history instead of just current message
research_plan=None,
report=None,
next_step="research_manager"
)
# # Create initial state using MessagesState
# initial_state = ResearchState(
# messages=[HumanMessage(content=message)],
# research_plan=None,
# report=None,
# next_step="research_manager"
# )
# Run workflow with optional tracing
config = {"callbacks": [self.tracer]} if self.tracer else {}
result = self.workflow.invoke(initial_state, config=config)
print("\n=== RESEARCH COMPLETED ===")
# Write the final report to a file
try:
with open("final_report.md", "w") as f:
f.write(result["messages"][-1].content)
print("Report saved to final_report.md")
except Exception as e:
print(f"Could not save report to file: {e}")
# Return the final report
return result["messages"][-1].content |