Field descriptions for DeltaV
The importance of describing your fields in Agent Functions
At registration of an Agent Function on the Agentverse (opens in a new tab)
, providing comprehensive information, especially in the description section, is paramount. A detailed Function description not only elucidates the functionality of the Agent Function but also aids users in grasping its essence. This clarity fosters effective interactions, empowering users to navigate the system with confidence, knowing precisely what to anticipate.
Similarly, the field description of the Agent Function needs to be detailed enough so to provide a clear picture of each data Model the Function requires for it to be executable for Objective execution. Hence, these descriptions serve as a guiding beacon for users, developers, and LLM alike. Each field explanation elucidates the purpose, usage, and expected inputs required by each Model class, streamlining interactions and facilitating execution. Enhanced understanding of field descriptions enables the AI Engine to interpret user requests accurately, thereby ensuring accurate Agent Functions execution.
For clarity, reinforcing key concepts could be done through repetition of keywords within the description. Such an approach minimizes ambiguity and mitigates the risk of misunderstandings. Whether for users or LLM, a clear and descriptive field description lay the foundation for smooth interactions and reliable Agent Function delivery and execution.
Consider the following Agent Function:
Providing a detailed field description for the num_rolls data Model is important to correctly execute the Function. A well written field description enhances the LLM understanding of the type of Function and the user objective requires to be executed. This would help in the accurate interpretation and execution of users' requests.
It is possible to call additional other Sub-Services from either Objective or other Sub-Services by specifying this within the field description itself. For additional information on Objective Service and Sub-Service, check the following resource
For a network of Objective Services and Sub-Services, please refer to the below examples.
Auto-description functionality
It is possible for you to generate auto-descriptions for your Functions when it comes to deploy them on Agentverse.
You can make use of this functionality by opening your Agent's details by clicking on the Agent's box within the list of Agents available in My Agents tab in Agentverse. Then, head over to the Deploy tab as shown below:
Here, you will need to either create a New Function or edit an existing one. This way, you will open your Agent's Function editor, and you will be able to provide/edit the description for what that Function does. It is here where you can make use of the auto-description functionality to fill in such field. At the moment, this functionality is available just for the Function Description, but it will be available for Fields description at a later stage.
Check out the following screenshot to have a better idea:
By clicking the Refine button, the AI Engine will automatically generate a description based on the Function details you have provided. You can either provide an initial prompt within the Description field to be then refined, or you can auto-generate a description from a nothing and the AI Engine will do this for you by itself.
Example 1: News Reading System
Primary Function Field Description
This Function helps user to read news of a specific type. This Function calls a Secondary function to generate news.
Remember to always provide a comprehensive description for triggering the Secondary Function in the field description so to ensure that the Secondary Function is always initiated.
The field description for the news data model describes the news that will be presented to the user. It should be mentioned that it should be always provided by triggering of Secondary Function. In our case, a good field description would be: Describes the news which will be generated from the "Generate News" secondary function. Always go for "Generate News" secondary function only never ask this field from user. All the news articles generated are presented as strings.
Secondary Function Field Description
This Function helps task to generate news and send it to task or another Secondary Function. Below, there is an example where the users ask for news category they want to read and provide news to Primary Function. For a better understanding, you can find the overall guide for this example here .
The field description for the category data model describes the category for which the user wants to read the news. It should be mentioned that it should be always provided by user. In our case, a good field description would be: Describes the category provided by user about which he wants to get news for. This should always be provided by user in all cases. This primary function responds to "Generate News" secondary function. This should be from options business, entertainment, general, health, science, sports, technology.
Remember to always include the secondary function trigger in the field description. This is very important to ensure that the secondary function is being called. When selecting one of multiple secondary functions, you need to use different names for each one of the functions. This way, we avoid confusion for the LLM.
Example 2: Hugging Face Text Classification Models
This Function helps user to choose a text classification model and make a query to that model. This task hugging face system goes for Secondary Functions to get make request to hugging face API which in turn trigger model list Secondary Functions to get list of model related to a search keyword. User can make query to their selected model.
Primary Function Field Description
The field description for the response describes the response to be provided to the user. The Primary Function always triggers secondary function Hugging Face Request. The ideal field description in this case will be like Describes the response to the user query. Always go for Hugging face request secondary function to get this field. Never ask this from user.
Secondary Function Field Description
Hugging Face Request
Hugging face request: This function has two fields model_id (to which query is to be made) and query (what request you want to make to the model). The model_id field again triggers Model List secondary function which asks user for search keyword to get most downloaded models related to that keyword.
Ideal field description for this example will be like:
model_id: Always go for model list secondary function. Never ask this field to user. query: Describes the query user wants to ask the model. Always ask this to user after model_id is given by model list secondary function.
Remember to always provide a comprehensive description in query field that first we should fetch model_id using secondary function before asking query to user.
Model List
Model List: This secondary function has field Search which searches for top 10 downloaded models related to that keyword from hugging face API and lets user selects from provided model id's.
Ideal field description for this example will be like:
Search: This is the type of model user wants to make query to. Always ask this to user. This always gives list of options to the user. Make user select one from these options.
Example 3: Local Business Finder
This Function helps user to find local businesses in any locality and specified category. This Agent Function returns list of 10 businesses to user and gives information about selected business by the user.
Primary Function Field Description
There are 3 fields in this primary function category (Type of business which user wants to search), city (City in which user wants to look around for business) and name (Name of the business which user wants to search).
Ideal field descriptions for this example will be like:
category: This describes business category provided for which user wants to search business. Ask this from user. city: always go for city finder secondary function to get user's city. Ask user if they want to use this city in yes/no question. ['yes': 'use this city', 'no': 'ask for city they want to search for']. name: Always go for secondary function business finder service's response. Never ask this from user.
Secondary Function Field Description
-
City Finder: This secondary function takes user location from DeltaV and looks for their current location. Below given is the ideal description for location.
location: This describes the coordinates of the city given in str(lat,long) format.
-
Business Finder: This secondary function takes category and city from the primary function and looks around for 10 business name in that area. Asks user to select one and then sends back to task as name field.
category: This is the category provided by the user for which they want to get businesses in primary function business details service, Use this category for task as well.
city: This is the city responded by city finder or one given by user.
Agent Function registration examples
For additional information and guidelines on how to register an Agent Function on the Agentverse and examples of descriptions, head over to our dedicated resources: