> ## Documentation Index
> Fetch the complete documentation index at: https://clearskies.cc/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Agents

> An agent is a reusable AI unit you build once and put to work — inside a workflow, and optionally behind its own command — all grounded in your Context Graph.

An agent is a reusable unit of AI you build once and use in more than one place. You give it a model and instructions, publish it, and then it's ready to do work: reason over a call, draft a recap, classify a deal, research an account. The same agent can run inside a [workflow](/docs/workflows/overview) as a step, and — when you give it a command — be called on directly by name. Build it once, use it wherever the work happens.

## Where an agent is used

* **Inside a workflow.** A workflow's [Run Agent step](/docs/workflows/actions/run-an-agent) runs a published agent over the workflow's data and hands the result to the steps that follow — so the workflow can summarize, decide, and write in a single run.
* **On its own command.** An agent can have a **command** so it can be called directly by name. The command is optional — an agent that only runs inside workflows doesn't need one. See [Build an agent](/docs/agents/building-an-agent) for the naming rules.

## Draft, then publish

Every agent starts as a **draft** when you create it. A draft is yours to shape and change; it isn't live yet. When it's ready, you **publish** it, and that's what makes it usable — a workflow can only reference a *published* agent, and a workflow can't publish while it points at an agent that's still a draft. So the order is always: build and publish the agent first, then wire it into a workflow. Continue editing later, and your changes stay in the draft until you publish again.

## Build it once, use it everywhere

Because an agent is defined once and referenced by id, the same agent can back several workflows and its own command at the same time. Update the agent, publish, and every place that uses it picks up the change. This is the heart of [building with Clearskies](/docs/building-with-clearskies/what-you-can-build): reusable pieces you assemble into the automation your team needs.

## In practice

You build a "call summarizer" agent, choose a model, and publish it. A **meeting ended** workflow references it in a Run Agent step to summarize each call and post the result to Slack. Later you give the same agent a command so a rep can also call it directly. One agent, two ways to use it — and one place to maintain it.

## Next steps

1. [Sign in to Clearskies](https://app.clearskies.cc/)
2. [Build an agent](/docs/agents/building-an-agent) — choose a model, optionally a command, and publish
3. [Use it in a workflow](/docs/agents/use-in-a-workflow) through the Run Agent step

Questions? We're here to help at [support@clearskies.cc](mailto:support@clearskies.cc)