Agent

The Agent Block

An agent is an AI entity capable of decision making. Given a goal, it undertakes a series of actions, observing and reasoning as it goes, until it achieves its intended outcome. Agents are powered by large language models and are prompted instructions that guide them in choosing which skills to deploy and how to react to the outcomes of those skills.

To understand more about how agents choose skills, read the section on Skill.

We have abstracted away the complexities of prompting and programming that make agents work, though we have left some prompting for you to customize. You also have the opportunity to provide context to the agent. An example of this would be:

You are an assistant at a clothing store.

Provided the examples above, the final prompt passed to the agent would be structured as something like this:

You are an assistant at a clothing store.

Use the following format:
Question: the input question you must answer
Thought: you should always think about what to do
Final Answer: the final answer to the original question

Begin!

Question: {input into the agent}

Agent Settings

Input

This prompts the agent of the goal you would like it to achieve. Instead of formatting the label to always be Input, we allow you to customize it. Customizing the label can help prompt the agent more clearly depending on the use case.

Thought

This section gives you the opportunity to prompt the agent with special instructions on how it should react to the outcome of a skill as well as instructions on how it should deploy a skill if at all. Leaving the instructions blank will provide default instructions to the agent.

Final Answer

This prompts the agent on how you would like it to respond with its final answer. An example instruction would be to instruct the agent to respond with "I do not know" if the agent is unsure of its answer. Another example would be to instruct the agent to respond concisely.

How agents work

An agent begins with being provided a prompt and an input. The prompt and input are then combined and used as an input into a language model. Next, the following sequence is repeated any number of times until the agent determines it has reached its goal:

1. The output of the language model is then parsed and a skill is chosen to be executed if necessary

2. Skill chain executes

3. The outcome of the skill and the initial prompt is passed into a language model

Finally, the agent concludes by outputting a final answer.