noah-read / agent /react_agent.py
yuki-zmstr's picture
Back to 4o
f2ed362
"""
ReAct Agent for Noah - A reasoning and acting agent that can:
1. Recommend books using OpenLibrary
2. Schedule reading time in Google Calendar
3. Create book reviews in Notion
4. Handle general conversation
"""
import json
import re
import os
import logging
from typing import Dict, List, Optional, Any
from langchain_core.messages import BaseMessage, HumanMessage, AIMessage, SystemMessage
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.tools import Tool
from langchain_openai import ChatOpenAI
from langchain_core.agents import AgentAction, AgentFinish
from langchain_core.output_parsers import PydanticOutputParser
from pydantic import BaseModel, Field
from dotenv import load_dotenv
from services.book_service import BookService
from services.calendar_service import CalendarService
from services.notion_service import NotionService
load_dotenv()
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class AgentDecision(BaseModel):
"""Model for agent's reasoning and action decision"""
thought: str = Field(description="The agent's reasoning about what to do")
action: str = Field(
description="The action to take: 'book_recommendation', 'schedule_reading', 'create_review', or 'FINAL_ANSWER'")
action_input: str = Field(description="The input for the chosen action")
class ReActAgent:
"""ReAct (Reasoning and Acting) Agent for Noah"""
def __init__(self):
self.llm = ChatOpenAI(model="gpt-4o", temperature=0.1)
self.book_service = BookService()
self.calendar_service = CalendarService()
self.notion_service = NotionService()
self.parser = PydanticOutputParser(pydantic_object=AgentDecision)
# Initialize tools
self.tools = self._create_tools()
self.tool_map = {tool.name: tool for tool in self.tools}
# ReAct prompt template
self.react_prompt = ChatPromptTemplate.from_messages([
SystemMessage(content=self._get_system_prompt()),
MessagesPlaceholder(variable_name="chat_history"),
("human", "{input}"),
("assistant", "{agent_scratchpad}")
])
# Conversation history
self.chat_history: List[BaseMessage] = []
def _get_system_prompt(self) -> str:
"""Get the system prompt for the ReAct agent"""
return """You are Noah, an AI reading assistant that helps users understand the meaning of words, discover books, schedule reading time, track their reading journey.
You have access to the following tools:
1. book_recommendation - Get book recommendations from OpenLibrary
2. schedule_reading - Schedule reading time in Google Calendar
3. create_review - Create book review/notes in Notion
Use the ReAct (Reasoning and Acting) framework:
1. THOUGHT: Think about what the user wants and what action to take
2. ACTION: Choose an appropriate tool or provide final answer
3. OBSERVATION: Analyze the tool result
4. Repeat if needed, or provide FINAL_ANSWER
For each response, provide your reasoning in JSON format:
{
"thought": "Your reasoning about what to do",
"action": "tool_name or FINAL_ANSWER",
"action_input": "input for the tool or final response"
}
Guidelines:
Explain meaning:
- If the user inputs only one word or phrase, you will explain its meaning
- your action will be FINAL_ANSWER, and action_input will be your response.
Book recommendations:
- You can search using genre or author
- The action input must be in JSON format.
- The schema is {'genre': str} or {'author': str}
- If you don't have enough information, ask the user for more information.
Scheduling:
- You will need the book title. The duration is optional.
- The action input must be in JSON format.
- The schema is {'book_title': str, 'duration': int}
- If the user uses a unit other than minutes, convert to minutes. e.g. 2 hours -> 120
Reviews:
- You will need the book title, author, rating, and review text.
- The action input must be in JSON format.
- The schema is {'book_title': str, 'author': str, 'review_text': str, 'rating': int}
- In the case you need to ask the user for more input, your action will be FINAL_ANSWER, and action_input will be your response.
- Always be friendly and enthusiastic about reading!
Available tools:
{tools}
{format_instructions}
"""
def _create_tools(self) -> List[Tool]:
"""Create the available tools for the agent"""
return [
Tool(
name="book_recommendation",
description="Get book recommendations based on genre, author, or preferences. Input should be a JSON string with 'genre', or 'author' fields",
func=self._recommend_books
),
Tool(
name="schedule_reading",
description="Schedule reading time in Google Calendar. Input should be a JSON string with 'book_title', and 'duration'",
func=self._schedule_reading
),
Tool(
name="create_review",
description="Create a book review or reading notes in Notion. Input should be a JSON string with 'book_title', 'author', 'review_text', 'rating' (1-5)",
func=self._create_review
),
]
def _recommend_books(self, input_str: str) -> str:
"""Tool for getting book recommendations"""
try:
params = json.loads(input_str)
query = params.get("query", "")
author = params.get("author", "")
genre = params.get("genre", "")
if genre: # search by genre, use OpenLibary
books = self.book_service.get_books_by_genre(genre)
elif author: # search by author, use GoogleBooks API
books = self.book_service.search_books_google(author=author)
elif query: # general keyword search, use GoogleBooks API
books = self.book_service.search_books(query=query)
else:
return "Please specify either a genre or search query for book recommendations."
if not books:
return f"No books found for '{genre or query}'. Try a different genre or search term."
result = "Here are some book recommendations:\n\n"
for i, book in enumerate(books[:5], 1):
result += f"{i}. **{book['title']}** by {book['author']}\n"
result += f" Rating: {book['rating']}\n"
result += f" {book['description'][:150]}...\n"
result += f" [More info]({book['link']})\n\n"
return result
except Exception as e:
return f"Error getting book recommendations: {str(e)}"
def _schedule_reading(self, input_str: str) -> str:
"""Tool for scheduling reading time"""
try:
params = json.loads(input_str)
book_title = params.get("book_title", "")
duration = params.get("duration", 30)
if not book_title:
return "Please specify a book title to schedule reading time for."
result = self.calendar_service.schedule_reading_session(
book_title=book_title,
duration=duration,
)
if result["success"]:
return f"✅ Scheduled {duration} minutes to read '{book_title}' at {result['scheduled_time']}"
else:
return f"❌ Failed to schedule reading time: {result['error']}"
except Exception as e:
return f"Error scheduling reading time: {str(e)}"
def _create_review(self, input_str: str) -> str:
"""Tool for creating book reviews in Notion"""
try:
params = json.loads(input_str)
book_title = params.get("book_title", "")
author = params.get("author", "")
review_text = params.get("review_text", "")
rating = params.get("rating", 5)
if not book_title:
return "Please specify a book title for the review."
result = self.notion_service.create_book_review(
title=book_title,
author=author,
review=review_text,
rating=rating
)
if result["success"]:
return f"✅ Created review for '{book_title}' in Notion: {result['page_url']}"
else:
return f"❌ Failed to create review: {result['error']}"
except Exception as e:
return f"Error creating review: {str(e)}"
def _parse_agent_decision(self, text: str) -> AgentDecision:
"""Parse the agent's decision from LLM output"""
# Try to extract JSON from the response
json_match = re.search(r'\{.*\}', text, re.DOTALL)
if json_match:
try:
decision_data = json.loads(json_match.group())
logger.info(f"Parsed agent decision: {decision_data}")
# Convert action_input to string if it's a dict/object
if isinstance(decision_data.get('action_input'), (dict, list)):
decision_data['action_input'] = json.dumps(
decision_data['action_input'])
elif not isinstance(decision_data.get('action_input'), str):
decision_data['action_input'] = str(
decision_data.get('action_input', ''))
return AgentDecision(**decision_data)
except Exception as e:
logger.error(f"Error parsing decision JSON: {e}")
raise
# Fallback parsing
if "FINAL_ANSWER" in text:
return AgentDecision(
thought="Providing final answer",
action="FINAL_ANSWER",
action_input=text
)
else:
return AgentDecision(
thought="Handling general conversation",
action="conversation",
action_input=text
)
def run(self, user_input: str) -> str:
"""Run the ReAct agent"""
self.chat_history.append(HumanMessage(content=user_input))
agent_scratchpad = ""
# Format the prompt
formatted_prompt = self.react_prompt.format(
tools="\n".join(
[f"- {tool.name}: {tool.description}" for tool in self.tools]),
format_instructions=self.parser.get_format_instructions(),
chat_history=self.chat_history[:-1],
input=user_input,
agent_scratchpad=agent_scratchpad
)
# Get LLM response
response = self.llm.invoke(
[HumanMessage(content=formatted_prompt)])
# Parse the decision
try:
decision = self._parse_agent_decision(response.content)
except Exception as e:
# Fallback to conversation
decision = AgentDecision(
thought=f"Error parsing decision: {e}",
action="conversation",
action_input=response.content
)
# Update scratchpad with thought
agent_scratchpad += f"\nThought: {decision.thought}"
agent_scratchpad += f"\nAction: {decision.action}"
agent_scratchpad += f"\nAction Input: {decision.action_input}"
# Check if final answer
if decision.action == "FINAL_ANSWER":
final_response = decision.action_input
self.chat_history.append(AIMessage(content=final_response))
return final_response
# Execute the tool
if decision.action in self.tool_map:
tool = self.tool_map[decision.action]
try:
logger.info(f"Executing tool: {decision.action}")
observation = tool.func(decision.action_input)
# Update scratchpad with observation
agent_scratchpad += f"\nObservation: {observation}\n"
self.chat_history.append(AIMessage(content=observation))
return observation
except Exception as e:
observation = f"Error: {str(e)}"
agent_scratchpad += f"\nObservation: {observation}\n"
else:
# Unknown action, treat as conversation
observation = self._handle_conversation(decision.action_input)
self.chat_history.append(AIMessage(content=observation))
return observation
def reset_conversation(self):
"""Reset the conversation history"""
self.chat_history = []
if __name__ == "__main__":
agent = ReActAgent()
agent.run("Schedule 45 minutes to read The Three Body Problem")