Mirasvit AI Assistant
v1.4.9

General Settings

image

OpenAI Secret Key

Navigate to Stores > Settings > Configuration > Mirasvit extensions > AI Assistant > General. Paste the API key into the field OpenAI Secret Key in order to use the artificial intellect in your store.

Note

The AI Assistant extension accesses ChatGPT via an API. You need to sign up for a ChatGPT account in order to obtain the OpenAI secret key.

Generate the OpenAI secret API key on page platform.openai.com/account/api-keys. Click Create new secret key and copy the key.

image

Using ChatGPT API costs money, which are billed from your ChatGPT account. The total pricing depends on the number of tokens generated by this language model. Refer to openai.com/api/pricing/ for more info on prices.

OpenAI Model

Choose a most suitable language models for you to utilize. AI Assistant works via various language models.

  • GPT-4o (gpt-4o) is improved GPT-4 Turbo model. It is 2 times faster and up to 50% cheaper than GPT-4 Turbo model. Also has improved capabilities in non-English languages and uses a new tokenizer which tokenizes non-English text more efficiently than GPT-4 Turbo. Maximal support input is 128 000 tokens. It was trained on data existing before Oct 2023.
  • GPT-4 Turbo (gpt-4-turbo-preview) is improved GPT-4 model featuring improved instruction following, JSON mode and more. Maximal support input is 128 000 tokens. It was trained on data existing before Apr 2023.
  • GPT-4 (gpt-4) is more capable for complex tasks and gives better results on large texts. Maximal support input is 8 192 tokens. It was trained on data existing before Sep 2021.
  • GPT-3.5 (gpt-3.5-turbo) is optimized for chat. It is considered by OpenAI as a most capable GPT-3.5 model, and its price is 1/10th the cost of text-davinci-003. Maximal support input is 16 385 tokens. It was trained on data existing before Sep 2021.
  • GPT-3 (text-davinci-003) is suited for any language task. It is configured for longer output of better quality and consistent instruction-following. Its input can be up to 4 097 tokens. Its trained on data before Jun 2021.