JADE: The First Multi-Model Chatbot

The world of Large Language Models (LLMs) is vast and exciting, with each model having unique strengths and weaknesses. However, this variety presents a challenge: using all available LLMs is practically impossible due to cost and complexity. Wouldn't it be incredible to have an easy way to experiment with different models, compare their responses, and even choose the best model for a specific task?

This is precisely why JADE was built. With a focus on simplicity, JADE eliminates unnecessary features like file or image uploads, allowing you to interact with a variety of LLMs. This streamlined approach unlocks the potential to compare models, leverage their individual strengths, and even mitigate biases through multi-message conversations.

Multi-Models

JADE is the first Multi-Model chatbot. The idea is to use multiple models within the same conversation. Here are the key points:

1. When asking a question, you can query multiple models and compare their responses to choose the best one.
2. The selected response can be used as the basis for the next message across all models.

For example, a response from GPT-4 Omni can be used by Claude Haiku in the next interaction.

This approach offers several benefits. First, it ensures you always have access to the best possible response by leveraging the strengths of different models. Second, it provides a more comprehensive understanding of a topic by considering various perspectives. Finally, using responses from one model as context for another can lead to more engaging and insightful conversations.

More information

• Variety of AI models from different providers.

  With JADE, you can easily switch between models like GPT 3.5 or 4o, Gemini, Llama, Mistral, Claude, and more. Even custom endpoint. This means you can choose the best model for for your specific needs, whether it's for general knowledge, creative writing, or technical expertise. Having access to multiple models allows you to take advantage of their unique strengths and weaknesses, ensuring you get the most accurate and relevant responses. (See all models available in the last section)
  
  

• Multiple models in a single conversation.

  You can ask a question and receive responses from several models at once, enabling you to compare their answers and choose the most suitable one. This feature is particularly useful for complex queries where different models might offer unique insights or solutions.
  
  

• Duplicate models in a single conversation.

  Hoz JADE work is that you can create custom bot. Each bot have a name, model, temperature and system prompt. You can create as many bot as you want and select as many to answer each question. An example is creating the same model that reponse in different language.
  
  

• Reduce Hallucination.

  AI models sometimes generate information that is inaccurate or misleading, a phenomenon known as "hallucination." By using multiple models, JADE reduces each model's bias. This ensures that the responses you receive are more reliable and trustworthy.
  
  

• Pay only for what you use.

  JADE use API, so you get access to free credits or tiers depending of the provider (see next section). This is particularly beneficial for users who may not need to use the chatbot extensively. Once the free credit use, you pay based on the length of you message and the response generated in tokens (a token is around 3 characters).
  
  JADE starts with a free tier that allows you to send up to 200 messages a month. For more intensive use, you can upgrade for just $0.95/month. So you can use Llama 70b for free forever if using JADE with a Groq Cloud account for example.
  
  

• All providers.

  Providers available:

  Models available:

  • OpenAI - OpenAI offer 5$ credits when creating an API account. Around 10 000 small question to GPT-4 Omni or 100 000 to GPT-3.5 Turbo.

    • GPT 4 Omni
    • GPT 4 Turbo
    • GPT 4
    • GPT 3.5 Turbo
  
  
  • Anthropic - Anthropic offer 5$ credits when creating an API account. Around 2 000 small question to Claude 3 Opus or 120 000 to Claude Haiku.

    • Claude 3 Opus
    • Claude 3 Sonnet
    • Claude 3 Haiku
  
  
  • Mistral - Mistral do not offer free credits.

    • Mixtral 8x22b
    • Mixtral 8x7b
    • Mistral 7b
    • Mistral Large
    • Mistral Small
    • Codestral
  
  
  • Groq - Groq offer a free tier with limit of tokens and request per minutes. The rate is plenty for a chatbot. 30 messages and between 6 000 and 30 000 tokens per minute. Per tokens coming soon.

    • Llama 3 70b
    • Llama 3 8b
    • Mixtral 8x7b
    • Gemma 7b
  
  
  • Google - Like Groq, Google offer a free tier with limit of tokens and request per minutes. The rate is plenty for a chatbot. 15 messages and 1 000 000 tokens per minute. Per tokens also available.

    • Gemini 1.5 pro
    • Gemini 1.5 flash
    • Gemini 1.0 pro
  
  
  • Perplexity - Perplexity do not offer a free tier or credits. Perplexity offer what they call 'online' models that can search online. So you can ask for the current weather for example. Those models have additional cost of 5$ per 1 000 requests.

    • Sonar Large
    • Sonar Large Online
    • Sonar Small
    • Sonar Small Online
    • Llama 70b
    • Llama 7b
    • Mixtral 8x7b
  
  
  • Hugging face - You can also use custom endpoints. I only tested hugging face but in theory, as long as the key is valid and it use the openai api, it should work. This part need some testing and improvement.