• Represents a specific context or purpose for accessing the platform
  • Has its own API key for authentication
  • Can be restricted to use only specific Apps (select from the list of Configured Apps) within the Project
  • Can have Custom Instructions for how to (and how not to) use Functions
  • Allows for a natural multi-agent architecture within a single Project

Agents are designed to support multi-agent systems. Each agent can have its own set of allowed Apps and Custom Instructions, enabling specialized behaviors. For simple use cases, you can use the default agent created for your Project.

​
Agent Level Access Control

For each App you configured, you can specify if you want to allow the agent to access that App.

​
Custom Instructions

You can specify a Custom Instruction per Function for the Agents to follow. e.g.,:

  • GMAIL__SEND_EMAIL: “Don’t send emails to people outside my organization”
  • BRAVE_SEARCH__WEB_SEARCH: “Only allowed to search the web for the following topics: […]”
  • GITHUB__STAR_REPOSITORY: “Don’t star repositories that are not related to AI”