| import os |
| from langchain_openai import ChatOpenAI |
| from langchain_core.messages import HumanMessage, AIMessage, BaseMessage |
| from langgraph.prebuilt import create_react_agent |
| from typing import List |
|
|
| from src.agent.tools import search_papers, compare_papers, generate_literature_review |
|
|
| SYSTEM_PROMPT = """You are ChatPaper, an expert AI research assistant. You have access to a database of research papers. |
| |
| CRITICAL RULES — NEVER BREAK THESE: |
| 1. You MUST call search_papers tool FIRST before answering ANY question. No exceptions. |
| 2. NEVER say you don't have enough information before searching. Always search first. |
| 3. NEVER ask the user for clarification before searching. Search with what you have. |
| 4. Base your answer ONLY on what the search_papers tool returns. |
| 5. If the tool returns no relevant results, then say the paper does not contain that information. |
| |
| Your tools: |
| - search_papers: search the indexed papers — call this IMMEDIATELY for every question |
| - compare_papers: compare topics across multiple papers |
| - generate_literature_review: write a literature review from the papers |
| |
| Answer format: |
| - Be concise and specific |
| - Always mention which paper and page your answer comes from |
| - Use bullet points for lists |
| """ |
|
|
|
|
| class ChatPaperAgent: |
|
|
| def __init__(self): |
| print("Initializing ChatPaper agent...") |
|
|
| self.llm = ChatOpenAI( |
| model="anthropic/claude-3-haiku", |
| api_key=os.getenv("OPENROUTER_API_KEY"), |
| base_url="https://openrouter.ai/api/v1", |
| temperature=0, |
| max_tokens=4096, |
| ) |
|
|
| self.tools = [ |
| search_papers, |
| compare_papers, |
| generate_literature_review, |
| ] |
|
|
| self.graph = create_react_agent( |
| model=self.llm, |
| tools=self.tools, |
| prompt=SYSTEM_PROMPT, |
| ) |
|
|
| self.conversation_history: List[BaseMessage] = [] |
|
|
| print("Agent ready!") |
|
|
| def chat(self, user_message: str) -> str: |
| |
|
|
| self.conversation_history.append( |
| HumanMessage(content=user_message) |
| ) |
|
|
| try: |
| result = self.graph.invoke( |
| {"messages": self.conversation_history} |
| ) |
|
|
| response_text = "" |
| for message in reversed(result["messages"]): |
| content = message.content |
|
|
| if isinstance(content, str): |
| if content.strip(): |
| response_text = content.strip() |
| break |
| elif isinstance(content, list): |
| text_parts = [] |
| for block in content: |
| if isinstance(block, dict): |
| if block.get("type") == "text" and block.get("text", "").strip(): |
| text_parts.append(block["text"].strip()) |
| elif isinstance(block, str) and block.strip(): |
| text_parts.append(block.strip()) |
| if text_parts: |
| response_text = " ".join(text_parts) |
| break |
|
|
| if not response_text: |
| response_text = "I could not generate a response. Please try again." |
|
|
| self.conversation_history.append( |
| AIMessage(content=response_text) |
| ) |
|
|
| return response_text |
|
|
| except Exception as e: |
| error_msg = f"Agent error: {str(e)}\n\nPlease try rephrasing your question." |
| return error_msg |
|
|
| def reset(self) -> None: |
| """ |
| Clear conversation history to start a fresh session. |
| Call this when the user clicks "New Conversation" or |
| when switching to a different set of papers. |
| """ |
| self.conversation_history = [] |
| print("Conversation reset") |
