Claire - The AI Assistant

Claire is a agentic AI that features persistent semantic memory, allowing it to maintain context across interactions. It engages users in natural-sounding conversations by leveraging large language models. Additionally, Claire can perform certain tasks with the set of tools it has been provided with.

Features

• Voice Interaction: Supports speech recognition and text-to-speech.
• Semantic Memory: Utilizes MongoDB Atlas Vector Search for semantic memory.
• Flight Search: Uses Amadeus API to find flights and closest airports.
• News Fetching: Retrieves the latest news by category and keyword.
• Trivia & Jokes: Engages users with fun facts and humor.
• Weather Updates: Provides real-time weather information.
• Wikipedia Lookup: Fetches brief Wikipedia summaries.
• Reminders & Notifications: Allows users to set reminders.
• Math Calculations: Evaluates mathematical expressions.
• Web Search: Uses Tavily Search for quick information lookup.
• Location Services: Determines user location based on IP.
• Frequency Count: Counts the frequency of words and characters in a text.

Technologies Used

• LangChain: For managing AI interactions and tools.
• OpenAI: For AI language model, text-to-speech, and vector embeddings.
• MongoDB Atlas: For semantic memory.
• SpeechRecognition: For voice input.
• Tavily Search: For quick information lookup.
• Amadeus API: Flight and airport search.
• NewsAPI: Fetching latest news articles.
• OpenWeatherMap: Providing weather updates.
• Wikipedia Library: Fetching Wikipedia summaries.
• Plyer: For notifications and reminders.
• Geocoder: Fetching user location.
• Open Trivia Database API: Fetching trivia questions.
• JokeAPI: Fetching jokes.

Installation

Prerequisites (Python 3.13.1)

• API keys for OpenAI, Tavily, Amadeus, OpenWeatherMap, and NewsAPI

• MongoDB Atlas URI (not localhost).

• Store the API Keys in a .env file as shown below

  OPENAI_API_KEY=YOUR_OPENAI_API_KEY
  TAVILY_API_KEY=YOUR_TAVILY_API_KEY
  AMADEUS_CLIENT_ID=YOUR_AMADEUS_CLIENT_ID
  AMADEUS_CLIENT_SECRET=YOUR_AMADEUS_CLIENT_SECRET
  OPENWEATHERMAP_API_KEY=YOUR_OPENWEATHERMAP_API_KEY
  NEWS_API_KEY=YOUR_NEWS_API_KEY
  MONGODB_URI=YOUR_MONGODB_ATLAS_URI

Note: All the API keys required, apart from OpenAI, have a free tier.

Setup

1. Clone the Repository

   git clone https://github.com/pratham-jaiswal/claire.git
   cd claire-ai

2. Install Dependencies

   pip install -r requirements.txt

3. Run the Assistant

   python main.py

4. Usage Options

   • You can choose to provide input via voice or text.
   • You can choose to receive output as voice, text, or both.

5. Model Configuration
   You can modify the code to use a different OpenAI model. See the Model Guide for available options.

   llm = ChatOpenAI(
       model="gpt-4o",
       temperature=0.5  # Controls response randomness
   )

   Note: If your budget allows, it is highly recommended to use GPT-4o or a more advanced model. While GPT-4o mini offers decent performance, GPT-4o provides enhanced capabilities and more accurate responses, which significantly improve interaction quality. Refer to the Pricing Guide for cost details.

6. Speech Configuration
   You can customize the code to use different voices, models, and audio formats. See the Text-to-Speech Guide.

   spoken_response = client.audio.speech.create(
       model="gpt-4o-mini-tts",  # 'gpt-4o-mini-tts' is costlier than 'tts-1' and 'tts-1-hd'
       voice="alloy",
       response_format="mp3",
       instructions="Use a warm and friendly tone",  # 'instructions' do not work with 'tts-1' or 'tts-1-hd'.
       input=text
   )

7. Embedding Configuration
   You can switch to a different embedding model. Refer to the OpenAI Embeddings Guide.

   embeddings = OpenAIEmbeddings(model="text-embedding-3-large")

   Or, use a different embedding provider:

   from langchain_community.embeddings import FakeEmbeddings
   
   embeddings = FakeEmbeddings(size=1352)

   For detailed information on integrating various model providers with LangChain, refer to the Integration Guide.

Usage

Once started, Claire will listen for voice commands or accept text input. Now simply interact with Claire as you would with a human assistant.

Example Commands

• What's my name?
• What's the weather like in my city?
• Find me a flight from Kolkata to Mumbai.
• Tell me a dark programming joke.
• Set a reminder in 30 minutes. (The code must be running for this to work)
• Give me the latest news on technology.
• Who was the founder of reddit and what happend to him?
• How many times does the letter 'o' appear in the word 'pneumonoultramicroscopicsilicovolcanoconiosis'?

Contributing

Please read CONTRIBUTING.md for the process of submitting pull requests to us.