Configuring an Agent
All agents are configured using a YAML configuration file. For a very simple agent that simply interacts with the user using a specified LLM, your agent config file might look something like:
apiVersion: skagents/v1
kind: Sequential
description: >
A simple chat agent
service_name: ChatBot
version: 0.1
input_type: BaseInput
spec:
agents:
- name: default
role: Default Agent
model: gpt-4o
system_prompt: >
You are a helpful assistant.
tasks:
- name: action_task
task_no: 1
description: Chat with user
instructions: >
Work with the user to assist them in whatever they need.
agent: default
An agent configuration file can contain the following elements:
- apiVersion - At present, this should always be
skagents/v1
- kind - The way in which an agent will execute its tasks. Currently only the
value
Sequential
is supported, which means, for each invocation, the agent will execute all tasks in the defined order. - description (optional) - A description of the agent
- service_name - The name of the agent (and thus its service). This, in
combination with the version will make up the agent's REST and streaming
endpoints. In this example, said endpoints would be
/ChatBot/0.1
/ChatBot/0.1/stream
- version - The version of the agent (see above)
- input_type - The payload format for requests to this agent (more on this later)
- output_type (optional - not shown) - The payload format for responses from this agent (more on this later)
- spec - Agent and task configuration
- agents - A list of agents that can be used by tasks
- name - The name of the agent
- role - The role/description of the agent
- model - The LLM model to use
- system_prompt - A system prompt for the agent
- tasks - A list of tasks to be executed by the agent
- name - The name of the task
- task_no - The order in which the task should be executed
- description - A description of the task
- instructions - Instructions for the task
- agent - The agent to use for the task (must match agent defined in the agents section)
Currently available models
The models currently available for agent configuration include:
- gpt-4o (Not for MSD usage)
- gpt-4o-mini (Not for MSD usage)