🚧 These docs are still under construction. Reach out on Discord if you’d like more information on anything about Agent Cloud.
Agents are like your own personal GPTs except so much more capable. They’re able to gather data from given datasources, use that data to pass onto each other and can either chat with a user regarding them or process the data automatically. But agents are best used together, they can be used together inside an app to achieve an overall larger goal. Think of an agent as a member of a team, which specific skills and a particular job to do.

Agent Components

Agents are compomised of a few components which will be briefly touched on here but more can be found on their respective pages.

1. System Message

The system message is the prompt used to initialise the agent. It tells the agent their role, what their skills are and what their goal is.
The better these prompts are, the better the output of the agent will be, if there are problems with your app such as unexpected outputs the first place to take a look is generally here.
Role
The agent’s role defines their function within the crew of agents. This specifies the kind of task the agent is best used for so that when the app is deciding which agent to use for a specific task it can call upon the role to see what it’s used for.
Here’s an example role for an agent:
Role for a crew of agents where the overall goal is to analyse a user’s PC requirements and reccomend products from that:
Agent 1: Product Researcher
Agent 2: Preference analyst
Agent 3: Product Finder
Goal
The goal is the individual objective that the agent aims to achieve. It guidess the agent’s decision making process and is used to prompt the agent as to their purpose and the task they must accomplish.
Goals for the above mentioned crew:
Product Researcher: Being the best at interpreting and analysing product preferences to surprise customers with products they didn’t know they needed.
Preference analyst: Being the best at analysing responses to questions to establish a pattern of product preferences
Product Finder: Being the best at combining product preferences and gathered products to give the customer the best value for money package of products that they need.
Backstory
The backstory provides context to the agent’s role and goal, enriching the interaction and collaboration dynamics. Think of the backstory as a detailed job description.
Example backstories for the above crew:
Product Researcher: You are the the most experienced researcher to is able to analyse preferences for certain types of products and use those preferences to find out what the customer likes, once you’ve determined this you find these products in the catalogue.
Preference analyst: You are the most experienced customer analyst, you’re able to ask questions to find out what kind of products the customer is looking for and you use the responses to these questions to generate a set of preferences that the customer has for products.
Product Finder: You’re the most efficient and accurate analyst at picking out products to combine them into a package for the customer

2. Tools

Tools are a set of capabilities or functions that the agent can use to perform tasks. They can range from web searching and data analysis to collaboration and delegation of tasks among agents. Agent Cloud comes with a range of built in tools, here are a few:
This tool takes a string query as input and fetches relevant pages from Wikipedia.
PUT A VIDEO OF THE TOOL IN ACTION

See more on the Tools page.

3. Datasources

Datasources are the heart of an Agent Cloud app, it’s what sets our agents apart from your generic models. Agents can gather data from datasources to produce a specific output. For example, you can export all of your product data (SKU, product number, description, etc…) into Agent Cloud. Agents can then be used to not only reccomend generic products to the user (“Use an exfoliating hand cream”), but they can reccomend specific products that are stocked by you (“Use an exfoliating hand cream, xyz brand is great and we have it in stock!”). This is where the real power of Agent Cloud can be found.
See the Datasources page for information on how to set up Datasources.

4. Model

Simply put, the model is the model the agent uses as it’s LLM, this may be OpenAi’s GPT, or Meta’s Llama. Our platform supports many LLMs from multiple vendors.
See more on the Models page

Advanced Features

The following are optional features that aren’t neccessary to create an agent or app but they can help with debugging agents or enhancing their capability.

1. Funciton Calling Model

TODO: THIS SECTION IS NOT COMPLETE

2. Verbose

TODO: THIS SECTION IS NOT COMPLETE