Agent Best Practices
Agent Prompting
Agents make use of the reasoning capabilities of Large Language Models to power their decision making.
The agent LLM takes in several inputs formatted as text at each step:
Task Description: The description off the task to perform
Tool Descriptions: Descriptions of each tool and when to use them
Inputs: The User's provided inputs to the task
History: A text representation of the previous actions the agent has taken and their outputs
Based on the inputs the agent will select a tool to use and generate the inputs to pass to the tool. Then the VectorShift platform will execute the tool and feed its results back to the agent in order to be ready for the next step
Detailed task and tool descriptions help the agent determine what to do at each step. Writing clear instructions for the agent will help better adapt it to your tasks
Agents are highly customizable to specific tasks. Choosing the right tools and describing their use clearly will determine the success of the agent.
Last updated