Skip to main content
All CollectionsAssistants
Configuring Advanced Model Settings
Configuring Advanced Model Settings

Enhance control over your Cassidy Assistant's responses

Olivia Garrahan avatar
Written by Olivia Garrahan
Updated this week

*This is an advanced feature. Reach out to our team and we'll help you set it up!

The three advanced settings are:

  1. Max Prompt Tokens

  2. Temperature

  3. Referencing items in instructions

Follow these steps to configure advanced model settings for your assistant:

  1. Navigate to the Assistants page: Click on the "Assistants" option in the left sidebar of your Cassidy dashboard.

  2. Open the Assistant settings: Locate the Assistant you want to edit and click the "..." icon in the top right corner of the Assistant's card. From the dropdown menu, select "Edit.":

  3. Navigate to Advanced Model Settings: From the Assistant settings expand the "Advanced Model Settings" section:

Max Prompt Tokens

The Max Prompt Tokens is the number of tokens that can be allocated for each chat message with your Assistant.

The prompt tokens are shared for both prompt and completion. This means the length of your prompt input and the length of the Assistant's output will be taken into consideration when deducting the tokens. More tokens means you can have a longer input/output character length.

Increasing the number of tokens also increases the Assistant's memory. If you have a long thread with the Assistant, increasing the Max Prompt Tokens will enable the Assistant to consider earlier messages and attached files. An Assistant with less Max Prompt Tokens will only be able to reference recent messages.

* You will only be able to set a value for Max Prompt Tokens based on the token limits for the Base Model you selected for your Assistant

To change the Max Prompt Tokens setting for your Assistant, click and drag the slider to adjust the value:

Alternatively, you can type in the text box to set an exact value for Max Prompt Tokens:

Temperature

Temperature is the parameter that controls the randomness of the results generated by a given AI model.

For non-creative tasks where you want the Assistant to strictly adhere to instructions, it is best to use a lower Temperature. You should use a higher Temperature for highly creative tasks, such as marketing ideas, where there is room for the Assistant to output dynamic results.

To change the Temperature setting for your Assistant, click and drag the slider to adjust the value:

Alternatively, you can type in the text box to set an exact value for Temperature:

Referencing items in instruction

If you have advanced instructions enabled for your account, you will be able to directly reference items in the instructions for your Assistant, such as documents and folders, similar to the in-chat functionality of referencing knowledge base items. Any item referenced in the instructions for your Assistant will be inline, meaning the entire contents of the file will be added to your Assistant's instructions. Additionally, the contents of the files you include in the instructions will be present in every chat message.

  1. Add a new instruction: To add a new instruction, begin by clicking the "Add Instruction" button under the Assistant's instructions.

  2. Use the "#" key to reference Knowledge Base items: In the instruction text box, type the "#" key to open a dropdown menu that allows you to search for and select items from your Knowledge Base.

  3. Search for relevant items: In the dropdown menu, start typing keywords related to the information you want to reference. Cassidy will display a list of matching Knowledge Base items based on your search terms.

  4. Select the desired item: Click on the Knowledge Base item you want to reference in your instruction. The item will be added to the text box as a clickable link.

  5. Add additional context to instruction (Optional): To specify exactly how and when the Assistant should refer to the referenced information, you can add additional context to the instruction.

  6. Ask questions about the referenced information: Once you've referenced a Knowledge Base item, you can ask the AI assistant questions related to that information. For example, if you've referenced a company's vacation policy, you can ask questions like:

    • "How many vacation days am I entitled to according to the handbook?"

    • "What is the process for requesting time off?"

    • "How is vacation time accrued?"

    The Assistant will provide answers based on the content of the referenced item and the context of your organization.
    ​

* It is important to consider token limits when including items in the instructions for your Assistant, as the complete contents of the included files will contribute to the token limit when interacting with your Assistant. You may need to adjust the Max Prompt Tokens to accommodate the size of the items included in the instructions

When you include an item in your instructions, it's important to remember that you are attaching the document to the chat permanently, and the Assistant will always be looking at the document in its entirety.

If you have the Q&A feature enabled for your Assistant, the Assistant automatically searches the Knowledge Base with every request, but it will search for relevant snippets in your knowledge base, depending on the question the Assistant is asked. The Q&A Assistant will still do a search across the Knowledge Base, even if the item is referenced in the instructions. If you do not have the Q&A feature enabled for the Assistant, then the Assistant will not be able to search across the Knowledge Base, and will only be able to access content explicitly referenced in the instructions.

Did this answer your question?