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

# Understanding Agents

> How Velatir's agents review every trace and decide what happens next.

## What Are Agents?

Agents are the heart of Velatir. They review every trace that flows through your organisation and decide whether it should be allowed, flagged, blocked, or escalated. You do not trigger them by hand. They work in real time, applying the configuration you have set.

Velatir has two agents, and they work together. A single trace can be reviewed by both at once, each from its own angle.

<CardGroup cols={2}>
  <Card title="Gatekeeper" icon="brick-wall" href="/agents/gatekeeper">
    Controls which AI services your organisation can use. It enforces your default policy and any explicit exceptions, allowing or blocking services as people work.
  </Card>

  <Card title="Data Protector" icon="shield-check" href="/agents/data-protector">
    Catches sensitive content before it leaves your environment, across credentials, personal data, and financial data.
  </Card>
</CardGroup>

## Roles

Every agent runs with a **role** that sets how much authority it has. The role decides what actually happens when an agent has a concern.

<AccordionGroup>
  <Accordion title="Observer" icon="eye">
    The agent reviews traces and flags its findings for review. It does not notify anyone or block anything.

    **Best for:** the first weeks of a rollout, lower-risk workspaces, and building a picture of your usage before you enforce anything.
  </Accordion>

  <Accordion title="Enforcer" icon="shield-check">
    The agent blocks traces that break your rules, or escalates them to a person for approval before they proceed.

    **Best for:** workspaces handling sensitive data, regulated work, and any case where a violation must be prevented.
  </Accordion>
</AccordionGroup>

## Outcomes

When an agent reviews a trace, the result is one of four outcomes. Which one you see depends on the agent's role and what it found.

| Outcome       | What it means                                             |
| ------------- | --------------------------------------------------------- |
| **Allowed**   | No concern. The trace proceeds.                           |
| **Flagged**   | An Observer noted a finding for review, without blocking. |
| **Blocked**   | An Enforcer stopped the trace.                            |
| **Escalated** | The trace was sent to a person for review or approval.    |

## How an Outcome Is Decided

<Steps>
  <Step title="Each agent reviews the trace" icon="shield-check">
    Both active agents review the trace at the same time, each recording its own assessment.
  </Step>

  <Step title="Roles decide the action" icon="scale">
    An Observer only flags. An Enforcer can block, or escalate for approval.
  </Step>

  <Step title="The most restrictive outcome wins" icon="gavel">
    If either agent blocks, the trace is blocked. If either escalates, it is escalated. The trace proceeds only when nothing stands in its way.
  </Step>
</Steps>

<Tip>
  Start both agents in Observer to learn what they catch, then promote to Enforcer where the risk justifies it. See [Configuring agents](/agents/configuring-agents).
</Tip>

***

<CardGroup cols={2}>
  <Card title="Instructions" icon="fingerprint" href="/agents/instructions">
    Teach an agent how to treat a specific scenario.
  </Card>

  <Card title="Assessments" icon="list-checks" href="/agents/assessments">
    Review every verdict an agent has reached.
  </Card>
</CardGroup>