> ## 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.

# Configuring Agents

> Create named agents, pick an organisation default, and assign them to workspaces.

## Named Agents, Then Assignment

You do not edit one Gatekeeper, one Data Protector, and one Coach for the whole organisation. You create **named agents** of each type — as many as you need — and then decide which one applies where.

A typical setup is one default for most teams, and a stricter (or lighter) agent assigned to the workspaces that need different treatment.

## Create Agents

Under **Agents**, each type has its own list. Create a new Gatekeeper, Data Protector, or Coach, give it a name, and configure it. You can have several of the same type: for example a Gatekeeper that allows broadly, and another that only allows an approved list.

Each agent is a full configuration. Gatekeeper holds the default rule and exceptions. Data Protector holds the role, categories, capabilities, and instructions. Coach holds which messages are on and what they say.

## Organisation Default

For each type, mark one agent as **Default**. That is the one that applies when a workspace does not pick a specific agent — and when you have no workspaces yet.

Set as default from the agent card on the Agents overview. Switching the default does not change workspaces that already have a specific agent assigned.

## Assign to Workspaces

Open a workspace to choose which agent of each type it uses.

| Choice            | Behaviour                                                                                                                         |
| ----------------- | --------------------------------------------------------------------------------------------------------------------------------- |
| **Inherit**       | Use the agent assigned on the nearest parent workspace. If nothing is assigned further up the tree, use the organisation default. |
| **A named agent** | Use that agent for this workspace, regardless of what sits above it.                                                              |

Most workspaces can stay on Inherit. Assign a specific agent only where the team needs different rules.

Workspaces nest, so you can set a Gatekeeper on Engineering and let Platform and Mobile inherit it, while Legal gets a different one.

There is no separate off switch per workspace. A type is uncovered only when no agent of that type exists, or the default has been deleted and nothing is assigned.

## Localisation

Agent notifications can reach people in their own language. Configure this under **Settings → Agents**.

* Set an organisation **default language**, or choose **User default** so each person sees notifications in the language of their browser or OS, falling back to the default when it is not one of the supported languages.
* A workspace can set its own language.
* Edit the wording of each notification per language on Coach and Gatekeeper, starting from built-in defaults. A live preview shows the notification as it appears in the browser.

Supported languages today: English, Danish, Dutch, German, Spanish, Italian, French, and Portuguese.

Notifications use these languages once people are on **version 1.10.0** or later of the Velatir browser extension.

## Sandbox Testing

Try changes before they reach production. Open the testing environment from the flask icon in the top navigation.

* Test a **single instruction** in isolation to check one rule.
* Dry-run a **Data Protector** configuration (role, categories, and instructions) against a sample prompt.

Testing uses the same rules your agents use, and nothing you test touches live traffic or real sessions.

## Rolling Out

A phased rollout reduces disruption and builds confidence.

<Steps>
  <Step title="Create your first agents" icon="plus">
    Start with one Gatekeeper and, if you have it, one Data Protector. Set both as Observer. Set each as the organisation default. Turn Coach on when you want people to see guidance in the moment.
  </Step>

  <Step title="Review and tune" icon="list-checks">
    Look through [Assessments](/agents/assessments) to understand your usage. Set your Gatekeeper policy, enable the Data Protector categories that matter, and add [instructions](/agents/instructions) for the specific cases.
  </Step>

  <Step title="Promote to Enforcer" icon="shield-check">
    Move the default Gatekeeper and Data Protector to Enforcer when the risk justifies it. Turn on [access requests](/agents/gatekeeper#access-requests) and [false-positive challenges](/agents/data-protector#false-positive-challenges) so a block is not a dead end.
  </Step>

  <Step title="Add more agents where teams differ" icon="layout-grid">
    Create a second agent of the same type for a stricter or lighter policy, and assign it on those workspaces. Leave everyone else on Inherit.
  </Step>
</Steps>

## Setup Checklist

| Step                                          | Description                                                                      |
| --------------------------------------------- | -------------------------------------------------------------------------------- |
| Create a default of each type                 | One Gatekeeper, one Data Protector (if enabled), one Coach, each marked Default. |
| Set notification language                     | Pick a default, or User default, under **Settings → Agents**.                    |
| Identify workspaces that need different rules | Sensitive teams, approved-only lists, or teams that should stay in Observer.     |
| Create extra agents and assign them           | Assign the named agent on those workspaces; leave the rest on Inherit.           |
| Monitor and adjust                            | Review assessments regularly and refine over time.                               |

***

<CardGroup cols={2}>
  <Card title="Understanding agents" icon="shield-check" href="/agents/understanding-agents">
    Roles, outcomes, and how verdicts are decided.
  </Card>

  <Card title="Instructions" icon="fingerprint" href="/agents/instructions">
    Add your own rules for specific scenarios.
  </Card>
</CardGroup>
