An agent is an AI agent that answers questions and performs tasks on the platform. It is not built and deployed on its own: an agent is a feature of a service component, and one service can provide more than one agent.
Agents, Services, and Projects
An agent belongs to the service that provides it, and that service belongs to a project:
-
A project is a versioned, deployable unit made of one or more components.
-
A service is a component of a project. A service can provide one or more agents.
-
An agent is a feature of a service. It is not versioned on its own: an agent carries the version of the project that ships its service.
You reach an agent through the service that provides it. An agent therefore has no endpoint of its own, and it is available only while the service deployment that hosts it is running.
Refer to Key Concepts for the definitions of project, component, deployment, and environment.
How an Agent Is Deployed
An agent is never deployed on its own. It reaches an environment as part of the project that ships its service:
-
A developer creates a service using the PAS command-line interface (CLI) and creates one or more agents in it.
-
The build pipeline publishes the project the service belongs to, as a version.
-
You deploy that version to an environment from the Projects Catalog. The services of the project start, and the agents they provide start with them.
Each deployment of the project produces one deployed agent per agent and environment. An agent that is deployed to several environments therefore exists several times, once per environment, and each of these deployed agents has its own status and its own usage figures.
The Agents view lists deployed agents only. An agent that a project declares but that is not deployed anywhere is not listed.
Refer to Deploying a Project for details on the deployment of a project.
The Status of an Agent
A deployed agent reports one of four states: Started, Starting, Stopped, or Failed. The status follows the service deployment that hosts the agent.
An agent never reports the status Multiple, which applies to a project deployment as a whole.
Refer to Status and Badge Reference for all status values of the Control Hub.
Agent Records
The platform tracks the extent to which individual agents are used: how often they were called upon and how many input and output tokens they consumed. The figures are cumulative since the agent was first deployed. They are recorded while agents are handling calls, so there may be a slight delay compared to a call that has just been made.
The Control Hub displays these figures in the columns Calls and Tokens of the Agents view. It does not report the model an agent runs on, and it does not derive a cost from the consumed tokens.
Refer to Working With the Agents View for reading the usage figures.
Limitations
The Agents view is a read-only overview. You do not create, start, stop, restart, or delete an agent on its own. An agent follows the service deployment that hosts it: it runs while that deployment runs, and it stops when that deployment stops. Refer to Operating a Deployment for the lifecycle actions of a deployment, and to Deploying a Project for details on how to deploy a project.
You configure an agent through the configuration of the deployment or of the service that provides it. Refer to Configuring a Deployment for details.