General settings

This section describes the General settings tab in your AI Assistant / Chatbot.

Instructions

This is where you define your assistant's personality, you tell it what it is and how to behave. A simple example would be: "You are a helpful AI assistant, providing answers to questions from users", which is also the default instructions for every new assistant you create.

You can tell your assistant how to behave, how to respond and define some boundaries. Modifying these instructions can give very different results and we encourage you to experiment with various instructions and see how your assistant changes its behavior.

We include these instructions as part of ChatGPT's "system message" in our prompt engineering strategy.

Model

We currently support two different models, GPT-3.5 Turbo and GPT-4o (omni). These have a few distinct differences:

• GPT-3.5 Turbo
  GPT-3.5 models can understand and generate natural language or code and are faster and more cost effective than GTP-4 models. Costs less credits per tokens used. Has a small context window. Recommended for lesser costs.
• GPT-4o
  GPT-4o (omni) can solve difficult problems with greater accuracy than GPT 3.5 models. GPT-4 models have broader general knowledge and more advanced reasoning capabilities. GPT-4 is more expensive, costs more credits per tokens used. Has a large context window. Recommended for best results.

Which one you choose depends a little on your use case. We suggest monitoring responses and see if the answers you are getting are good enough using GPT-3.5 Turbo, if not, then you can try to switch it up to GPT-4o. New assistants use GPT-3.5 Turbo by default. For best results, we recommend using GPT-4o.

Read more about different OpenAI models.

Context window management

Context window management refers to how many tokens your assistant should use for context and output. Different AI models (GPT-3.5 Turbo and GPT-4o etc.) have different limits when it comes to context windows. GPT-4o has a much larger, possible, context window than GPT-3.5 Turbo.

You can use context windows to control costs, but if you do so, it can also have a negative effect on quality of responses. Some times queries can even fail, when the assistant tries to go beyond your set limits. We strongly recommend having these setting set to auto.

These settings only sets a limit on the max size context window, but in most cases, the model does not use the whole context window, especially when using GPT-4o.

Input tokens context window

This controls the max value tokens your assistant should use for input context. Input context can be your own data, files, web pages etc, but also the content of your current conversation/thread. A larger input context provides better results, since your assistant have more context to work with.

Ouput tokens context window

This controls the max value tokens your assistant should use for output. Output in this context is simply the answer coming from your assistant. Larger context window, can provide better and longer answers.

Code interpreter

Code interpreter is a tool that enables the assistant to write and run code in a sandboxed environment. This tool can process files with diverse data and formatting, and generate files such as graphs.

Without this enabled, your assistant can still understand and write code, but it cannot create files or run the code and iterate on it. Having this enabled may increase costs, but it is recommended to keep this enabled, to improve performance.

Chatbot functionality / visibility

Your newly created AI Assistants are by default set to "private", meaning only you can access them, either by interacting with them directly in our dashboard, or by using integrations.

If you want to embed your AI Assistant on other sites, like a chatbot, you need to make it public (you can still set some restrictions on where to access it). Enable the "Chatbot functionality" in your settings to make your assistant public and to get access to all the different chatbot settings (like appearance, security, embedding, insights etc.).

Creativity

With this setting you have some control over how your assistant will respond when it comes to creativity. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.