# Agents

An Agent is a reusable specialist with its own instructions, tools, and shared resources. People can work with it through conversations; an application with permission can call it through an authenticated API. Its setup lasts beyond any one conversation.

## When an Agent fits

Use an Agent when the same kind of work should be available to people or applications with its own configuration. A [Chat](/docs/chats/overview) remains a durable place for ongoing conversational work. For an application with its own interface and business logic, use a [Project](/docs/projects/overview).

An Agent can be associated with a Project or Site for organization. That association does not grant application access, credentials, or Site release authority. Read [access](/docs/agents/access) before connecting an application. Start with the [Agent quickstart](/docs/agents/quickstart).

## Example: a report reviewer

A report-review Agent can use standing instructions, a review Skill, and authorized reference Files to assess incoming reports. People can run it in conversations; an application needs an explicit grant to invoke it.

Define the review outcome before adding integrations. Start with [Quickstart](/docs/agents/quickstart), then use [Configuration](/docs/agents/configuration) to establish the method and [Access](/docs/agents/access) to decide who can run it.

A [Workflow](/docs/workflows/steps) can assign this Agent a task in a fresh conversation.

## Next steps

- [**Access** Choose who or what may use the Agent.](/docs/agents/access)
- [**Configuration** Define the specialist's method and reference material.](/docs/agents/configuration)
- [**Execution** Review conversations and recurring work.](/docs/agents/execution)
- [**Lifecycle** Update, reuse, or retire it deliberately.](/docs/agents/lifecycle)