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

# Intents

> Use intents as AI playbooks that tell Remark agents when and how to handle specific shopper requests.

Intents are playbooks for Remark agents. They define which shopper situations should use a playbook, what the agent should do, and which [actions](/dashboard/actions) the agent can use.

Use intents for repeatable processes such as product guidance, returns, order status, warranty questions, shipping policy, or escalation to a human.

## How intents work

Each intent has four main parts:

| Part              | What it controls                                     |
| ----------------- | ---------------------------------------------------- |
| Name              | The scenario the intent covers                       |
| Matching criteria | When the agent should use the intent                 |
| Instructions      | The step-by-step playbook the agent follows          |
| Actions           | Tools the agent can use while following the playbook |

An intent can be enabled for live conversations, enabled only in the playground, or disabled while you edit it.

## Create an intent

1. Open **Intents** in the Remark dashboard.
2. Select **Add intent**.
3. Add an intent name that describes the shopper scenario.
4. Write matching criteria that explain when the intent should activate.
5. Write instructions for how the agent should handle the situation.
6. Add any actions the agent needs.
7. Enable the intent in the playground first so you can test it.
8. Enable it for live channels after the behavior looks correct.

<Tip>
  Start new intents in playground mode. Test the playbook with realistic shopper messages before enabling it for live conversations.
</Tip>

## Create intents with Remark Copilot

Remark Copilot can help create intents for you. Use Copilot when you want to draft a new playbook from a goal, support scenario, or existing customer question.

Copilot can help turn a request such as "create an intent for returns and exchanges" into a draft with matching criteria, instructions, and suggested actions. Review the draft before publishing it, especially any action setup, policy language, channel targeting, and live/playground settings.

After Copilot creates or edits an intent, test it in the playground before enabling it for live conversations.

## Write matching criteria

Matching criteria tell the agent when this intent applies. Describe the shopper's situation, not just keywords.

Good matching criteria are specific about:

* What the shopper is asking for
* What edge cases should still match
* What similar requests should not match

For example:

```txt theme={null}
Use this intent when the shopper wants to start a return or exchange for an order they already placed.

Do not use this intent when the shopper is asking about the return policy before buying, unless they also want to start a return for an existing order.
```

If two intents have similar matching criteria, make the boundary clear. Overlapping criteria can cause the agent to choose the wrong playbook.

## Write durable instructions

Instructions tell the agent what to do after the intent is selected. Treat them as process guidance, not a place to store facts that change often.

Avoid hardcoding:

* Current prices
* Active discount codes
* Stock levels
* Temporary shipping cutoffs
* Product lists that change frequently

Instead, tell the agent how to look up current information using knowledge, product, order, or integration actions.

<Warning>
  Avoid absolute words like "always", "never", and "must" unless the rule is truly absolute. These words can make the agent too rigid in edge cases.
</Warning>

## Give recourse

When instructions tell the agent not to do something, also tell it what to do instead.

Less useful:

```txt theme={null}
Do not make up return policy details.
```

More useful:

```txt theme={null}
Do not make up return policy details. If the policy is unclear, search the knowledge base. If the answer is still unclear, explain that the team needs to confirm and escalate to support.
```

## Add actions

[Actions](/dashboard/actions) are the tools the agent can use while following an intent. Examples include searching knowledge, looking up products, checking an order, creating a ticket, or using an integration-specific action.

To add an action:

1. Open the intent.
2. In the **Actions** section, select **Add an action**.
3. Choose the action the agent should be allowed to use.
4. Configure any required fields.
5. Reference the action in the instructions when you need the agent to use it at a specific step.

Available actions depend on your Remark setup and connected integrations. For example, Loop Returns actions are only available after Loop Returns is connected.

## Reference actions in instructions

You can write instructions in plain language, or you can reference a specific action when the exact tool matters.

Plain language is best for general guidance:

```txt theme={null}
Search the knowledge base for the latest return policy before answering.
```

Use an action reference when the agent needs to run a specific configured action:

```txt theme={null}
After confirming the order number and verification detail, use @LoopReturns.CREATE_RETURN_LINK to create the return link.
```

<Note>
  An action reference only works if that action is also added to the intent's Actions section.
</Note>

## Configure action fields

Some actions need fields. A field can be filled by the conversation or set to a constant value.

Use conversation-filled fields when the agent should collect or infer the value from the shopper conversation. Use constant values when the action should use the same value every time the intent runs.

For actions that create cards in the conversation, you can choose whether the generated card should be shown.

## Channel targeting

If your workspace has multiple conversation channels, an intent can be enabled for all channels or limited to specific channels.

Use channel targeting when the same shopper request should behave differently in chat, email, SMS, or helpdesk channels. Keep the shared process in the main instructions, then add channel-specific guidance only where behavior actually differs.

## Test and publish

Use the playground to test intents before they go live.

1. Enable the intent in playground.
2. Run realistic shopper messages that should match the intent.
3. Check whether the right intent was selected.
4. Check whether the agent followed the instructions and used the right actions.
5. Update matching criteria or instructions if the result is wrong.
6. Enable the intent for live conversations when the behavior is ready.

## Edit or archive an intent

Open an existing intent to update its name, matching criteria, instructions, actions, playground status, live status, or channel targeting.

Archive an intent when it should no longer be used. Archived intents are removed from normal intent lists and are not available to the agent.

## Common intent set

Most ecommerce brands should consider intents for:

* Product guidance
* Order status and tracking
* Order changes or cancellations
* Returns and exchanges
* Warranty or repairs
* Shipping and delivery policy
* General FAQ or account help
* Escalation to a human

Add category-specific product guidance intents when different product categories need different questions, research steps, or recommendation logic.

<Note>
  Need help planning or testing intents? Contact `support@remark.ai`.
</Note>