Skip to content

Latest commit

 

History

History
211 lines (169 loc) · 8.53 KB

multion.mdx

File metadata and controls

211 lines (169 loc) · 8.53 KB
title
MultiOn

Build a personal browser agent that remembers user preferences and automates web tasks. It integrates Mem0 for memory management with MultiOn for executing browser actions, enabling personalized and efficient web interactions.

Overview

In this guide, we'll explore two examples of creating Browser-based AI Agents:

  1. An agent that searches arxiv.org for research papers relevant to user's research interests.
  2. A travel agent that provides personalized travel information based on user preferences. Refer to the notebook for detailed code.

Setup and Configuration

Install necessary libraries:

pip install mem0ai multion openai

First, we'll import the necessary libraries and set up our configurations.

import os
from mem0 import Memory, MemoryClient
from multion.client import MultiOn
from openai import OpenAI

# Configuration
OPENAI_API_KEY = 'sk-xxx'  # Replace with your actual OpenAI API key
MULTION_API_KEY = 'your-multion-key'  # Replace with your actual MultiOn API key
MEM0_API_KEY = 'your-mem0-key'  # Replace with your actual Mem0 API key
USER_ID = "your-user-id"

# Set up OpenAI API key
os.environ['OPENAI_API_KEY'] = OPENAI_API_KEY

# Initialize Mem0 and MultiOn
memory = Memory()  # For local usage
memory_client = MemoryClient(api_key=MEM0_API_KEY)  # For API usage
multion = MultiOn(api_key=MULTION_API_KEY)

Example 1: Research Paper Search Agent

Add memories to Mem0

Define user data and add it to Mem0.

USER_DATA = """
About me
- I'm Deshraj Yadav, Co-founder and CTO at Mem0, interested in AI and ML Infrastructure.
- Previously, I was a Senior Autopilot Engineer at Tesla, leading the AI Platform for Autopilot.
- I built EvalAI at Georgia Tech, an open-source platform for evaluating ML algorithms.
- Outside of work, I enjoy playing cricket in two leagues in the San Francisco.
"""

memory.add(USER_DATA, user_id=USER_ID)
print("User data added to memory.")

Retrieving Relevant Memories

Define search command and retrieve relevant memories from Mem0.

command = "Find papers on arxiv that I should read based on my interests."

relevant_memories = memory.search(command, user_id=USER_ID, limit=3)
relevant_memories_text = '\n'.join(mem['text'] for mem in relevant_memories)
print(f"Relevant memories:")
print(relevant_memories_text)

Browsing arXiv

Use MultiOn to browse arXiv based on the command and relevant memories.

prompt = f"{command}\n My past memories: {relevant_memories_text}"
browse_result = multion.browse(cmd=prompt, url="https://arxiv.org/")
print(browse_result)

Example 2: Travel Agent

Get Travel Information

Add conversation to Mem0 and create a function to get travel information based on user's question and optionally their preferences from memory.

```python Code def get_travel_info(question, use_memory=True): if use_memory: previous_memories = memory_client.search(question, user_id=USER_ID) relevant_memories_text = "" if previous_memories: print("Using previous memories to enhance the search...") relevant_memories_text = '\n'.join(mem["memory"] for mem in previous_memories) 
    command = "Find travel information based on my interests:"
    prompt = f"{command}\n Question: {question} \n My preferences: {relevant_memories_text}"
else:
    command = "Find travel information based on my interests:"
    prompt = f"{command}\n Question: {question}"

print("Searching for travel information...")    
browse_result = multion.browse(cmd=prompt)
return browse_result.message

Example usage

question = "Show me flight details for it." answer_without_memory = get_travel_info(question, use_memory=False) answer_with_memory = get_travel_info(question, use_memory=True)

print("Answer without memory:", answer_without_memory) print("Answer with memory:", answer_with_memory)

Another example

question = "What is the best place to eat there?" answer_without_memory = get_travel_info(question, use_memory=False) answer_with_memory = get_travel_info(question, use_memory=True)

print("Answer without memory:", answer_without_memory) print("Answer with memory:", answer_with_memory)


```json Conversation
# Add conversation to Mem0
conversation = [
    {
        "role": "user",
        "content": "What are the best travel destinations in the world?"
    },
    {
        "role": "assistant",
        "content": "Could you please specify your interests or the type of travel information you are looking for? This will help me find the most relevant information for you."
    },
    {
        "role": "user",
        "content": "Sure, I want to travel to San Francisco."
    },
    {
        "role": "assistant",
        "content": """
        Based on the information gathered from TripAdvisor, here are some popular attractions, activities, and travel tips for San Francisco: \

        1. **Golden Gate Bridge**: A must-see iconic landmark. \
        2. **Alcatraz Island**: Famous former prison offering tours. \
        3. **Fisherman's Wharf**: Popular tourist area with shops, restaurants, and sea lions. \
        4. **Chinatown**: The largest Chinatown outside of Asia. \
        5. **Golden Gate Park**: Large urban park with gardens, museums, and recreational activities. \
        6. **Cable Cars**: Historic streetcars offering a unique way to see the city. \
        7. **Exploratorium**: Interactive science museum. \
        8. **San Francisco Museum of Modern Art (SFMOMA)**: Modern and contemporary art museum. \
        9. **Lombard Street**: Known for its steep, one-block section with eight hairpin turns. \
        10. **Union Square**: Major shopping and cultural hub. \

        Travel Tips: \
        - **Weather**: San Francisco has a mild climate, but it can be foggy and windy. Dress in layers. \
        - **Transportation**: Use public transportation like BART, Muni, and cable cars to get around. \
        - **Safety**: Be aware of your surroundings, especially in crowded tourist areas. \
        - **Dining**: Try local specialities like sourdough bread, seafood, and Mission-style burritos. \
        """
    },
    {
        "role": "user",
        "content": "Show me hotels around Golden Gate Bridge."
    },
    {
        "role": "assistant",
        "content": """The search results for hotels around Golden Gate Bridge in San Francisco include: \

        1. Hilton Hotels In San Francisco - Hotel Near Fishermans Wharf (hilton.com) \
        2. The 10 Closest Hotels to Golden Gate Bridge (tripadvisor.com) \
        3. Hotels near Golden Gate Bridge (expedia.com) \
        4. Hotels near Golden Gate Bridge (hotels.com) \
        5. Holiday Inn Express & Suites San Francisco Fishermans Wharf, an IHG Hotel $146 (1.8K) 3-star hotel Golden Gate Bridge • 3.5 mi DEAL 19% less than usual \
        6. Holiday Inn San Francisco-Golden Gateway, an IHG Hotel $151 (3.5K) 3-star hotel Golden Gate Bridge • 3.7 mi Casual hotel with dining, a bar & a pool \
        7. Hotel Zephyr San Francisco $159 (3.8K) 4-star hotel Golden Gate Bridge • 3.7 mi Nautical-themed lodging with bay views \
        8. Lodge at the Presidio \
        9. The Inn Above Tide \
        10. Cavallo Point \
        11. Casa Madrona Hotel and Spa \
        12. Cow Hollow Inn and Suites \
        13. Samesun San Francisco \
        14. Inn on Broadway \
        15. Coventry Motor Inn \
        16. HI San Francisco Fisherman's Wharf Hostel \
        17. Loews Regency San Francisco Hotel \
        18. Fairmont Heritage Place Ghirardelli Square \
        19. Hotel Drisco Pacific Heights \
        20. Travelodge by Wyndham Presidio San Francisco \
        """
    }
]

Conclusion

By integrating Mem0 with MultiOn, you've created personalized browser agents that remember user preferences and automate web tasks. The first example demonstrates a research-focused agent, while the second example shows a travel agent capable of providing personalized recommendations.

These examples illustrate how combining memory management with web browsing capabilities can create powerful, context-aware AI agents for various applications.

Help