> For the complete documentation index, see [llms.txt](https://docs.rox.com/development/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rox.com/development/product/custom-agents/what-is-a-custom-agent.md).

# What is a custom agent?

A custom agent is how you tell Rox:

**when this happens, do this, and deliver the result here.**

A custom agent takes work that used to wait on a person, someone remembering, someone checking, someone following up, someone pulling the pieces together, and sets it in motion automatically.

A meeting appears on tomorrow’s calendar, and Rox starts building the brief.

A meeting ends, and the follow-up starts writing itself.

A deal changes, a signal appears, a webhook arrives, and Rox moves.

Every agent has three parts:

* a **trigger**
* a set of instructions
* an **output**

#### Trigger

The moment that starts the motion.

A trigger tells Rox when it is time to run. Agents can start from a schedule, a time-based entity event, a transcript completing, a deal event, a webhook, or a Clever column cell generation event.

Examples:

* an agent that runs every Friday at 4 PM
* an agent that runs before or after a timed event
* an agent that runs when a meeting transcript is completed
* an agent that runs when a deal changes
* an agent that runs when a webhook is called
* an agent that runs when a Clever column cell is generated

#### Instructions

The work Rox does next.

Instructions are the natural language steps used by the agent to execute certain actions. Rox can pull context, analyze activity, generate content, update records, send messages, transform data, and store variables.

Examples:

* retrieve account, contact, or meeting context
* read transcripts and emails
* generate a brief, summary, or draft
* send a Slack message or email
* update a record in Rox
* save and reuse variables across different agent runs

#### Output

The result Rox creates, sends, updates, or sets into motion.

Agents can produce inbox and artifact outputs including:

* text
* html
* email
* slides
* file

That might look like:

* a prep brief in your inbox
* a formatted HTML report
* a ready-to-review email draft
* a slide deck
* a downloadable file

So when you hear “agent” think:

**signal in, work happens, result out.**

That is what an agent is.

A repeatable piece of work, turned into automatic execution.

A best practice, no longer left to chance.

\ <br>