> For the complete documentation index, see [llms.txt](https://atlas.anywhere.now/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://atlas.anywhere.now/channels-integrations/dialogue-ai-assist/manage-dialogue-ai-assist-assistants.md).

# Manage Dialogue AI Assist assistants

Create, configure, test, and monitor Dialogue AI Assist assistants in Deepdesk.

Assistant Editors configure the assistants that support agents with Dialogue AI Assist in WebChat and voice calls. Use the Deepdesk admin portal to configure assistant behavior, presentation, knowledge, integration, and run control.

## Access the Deepdesk admin portal

To access the Deepdesk admin portal, go to:

```
https://[account].deepdesk.com
```

The exact URL is unique for the organization. For the exact URL, contact your AnywhereNow representative.

To open the assistant area, select **Agentic AI** > **Assistants**.

## Assistant Editor areas

The Deepdesk admin portal contains several menu groups. Assistant Editor tasks use these areas:

| Area                                    | Purpose                                                                      |
| --------------------------------------- | ---------------------------------------------------------------------------- |
| **Agentic AI** > **Assistants**         | Create, configure, test, monitor, and assign assistants.                     |
| **Agentic AI** > **Settings**           | Configure assistant connectors, routes, and groups used by assistants.       |
| **Admin Settings** > **Profile**        | Assign assistants to profiles for event-based, knowledge, and on-demand use. |
| **Admin Settings** > **Knowledge Base** | Create and process knowledge sources that assistants can use.                |

## Review the assistant inventory

**Agentic AI** > **Assistants** shows the assistant inventory for the environment. Use **Add assistant** to create an assistant. Use search, **All Labels**, and the date range filter, for example **Last 30 days**, to find or filter assistants.

The page shows summary metrics for **Requests**, **Tokens**, and **Cost**. The assistant table contains:

* **Name**
* **Feedback score**
* **Errors**
* **Requests**
* **Model Cost**
* **Token Usage**

Rows are paginated.

## Review assistant tabs

The assistant editor contains these tabs:

| Tab             | Purpose                                                                                                |
| --------------- | ------------------------------------------------------------------------------------------------------ |
| **Insights**    | Review assistant usage and performance, including requests, rating, cost, date ranges, and CSV export. |
| **Behavior**    | Configure prompt, tools, presentation, knowledge, and runtime settings.                                |
| **Integration** | Configure whether another assistant can call this assistant as a tool.                                 |
| **Run Control** | Configure groups, run conditions, fallback assistant, and trigger visibility.                          |
| **Feedback**    | Review user feedback, assistant responses, user, and conversation ID.                                  |
| **Activity**    | Review assistant runs, status, duration, tokens, and output.                                           |
| **Experiments** | Add and review assistant experiments.                                                                  |

{% hint style="info" %}
Because of the data retention policy, experiment files are automatically deleted after the designated retention period and will no longer be accessible.
{% endhint %}

## Test an assistant

Use the test panel to try an assistant before relying on it in production workflows. The side panel is titled with the assistant name and shows **Start a conversation with \[assistant name]**. Use the input field with placeholder **Type your message...** to test the assistant response.

Before you enable an assistant for the whole team, test it in Deepdesk and ask a pilot agent to try it in WebAgent or on calls.

## Create and name assistants

When you create or edit an assistant, choose a clear name and a stable code.

* **Name** helps Assistant Editors recognize the assistant in the portal.
* **Code** identifies the assistant across assistant configuration and should remain stable after the assistant is used by assignments or other assistants.

To create an assistant:

1. Select **Add assistant**.
2. Enter **Name**.
3. Review the prefilled **Code**.
4. Select the code edit button if you need to change the generated code.
5. Enter **Instructions**.
6. Select **Save**.

{% hint style="warning" %}
The assistant code must be unique. Changing a code after the assistant is used by profiles, routes, or other assistants can break references.
{% endhint %}

## Configure assistant behavior

Use the **Behavior** tab to configure how an assistant responds and which supporting resources it can use.

### Prompt

Use **Prompt** to define what the assistant must do, how it should interpret available conversation context, when it should call tools, and how it should respond. The prompt text area is labelled **Instructions \***.

Good prompts are specific about:

* the assistant task;
* the expected input;
* the expected output;
* when tools should be used;
* what the assistant should do when required information is missing.

<details>

<summary>Example: handover summary and notification pattern</summary>

Use this pattern when an assistant should create a compact handover summary for an agent joining or taking over an ongoing interaction.

```
# Task
You are a helpful assistant supporting a customer service agent during an interaction.
Use the transcript and available metadata to create a compact handover summary for the agent.

# Output
Send a short notification that includes:
- the customer's reason for contact
- the most important facts
- actions already taken
- open questions or next steps

Do not include call-to-action buttons.
If the transcript does not contain enough information, say what is missing instead of guessing.
```

</details>

<details>

<summary>Example: sentiment insight pattern</summary>

Use this pattern when an assistant should classify the tone of the conversation and provide a short explanation.

```
# Task
Analyze the transcript and classify the customer's sentiment as positive, neutral, or negative.

# Output
Return one short sentence with the sentiment and the reason for the classification.

# Example
Input: The customer says the issue is resolved and thanks the agent for the help.
Output: Positive — the customer confirms the issue is resolved and expresses appreciation.
```

</details>

<details>

<summary>Example: Knowledge Base answer pattern</summary>

Use this pattern when an assistant must answer from a connected Knowledge Base.

```
# Task
Answer the agent's question using the connected knowledge source.

# Rules
Use the knowledge source as the source of truth.
If the answer is not available in the source, say that the answer is not available.
Include source information when it is available.
Keep the answer concise and agent-facing.
```

</details>

### Tools

Use **Tools** to configure the tools that are available to the assistant in the environment. Select **Add** to assign tools. If no tools are assigned, the section shows **No tools assigned**. Assigned tools appear by name and can be removed.

In **Assign tools to assistants**, choose the tools the assistant can use for its tasks. The assistant only works with the tools you select. Use **Type to search** to filter tools, then select **Assign** or **Cancel**.

Tools are grouped into these categories:

* **System tools**: Built-in tools visible in the environment.
* **Assistants**: Other assistants configured under **Agentic AI** > **Assistants**. Select an assistant here when this assistant should call it as a tool.
* **Microsoft Copilot**: Copilot tools configured through **Agentic AI** > **Settings** > **Connectors**.
* **APIs**: Tools generated from Assistant Connectors configured under **Agentic AI** > **Settings** > **Connectors**.

System tool examples include `answer_knowledge_assist_question`, `call_api`, `call_knowledge_assist`, `call_url`, `get_current_time`, `retrieve_knowledge`, `send_call_to_action`, `send_user_notification`, `suggest_as_reply`, `validate`, and `write_to_memory`.

If an assistant calls another assistant, include that assistant in the available tools and configure the called assistant on the **Integration** tab so it is available as a tool.

<details>

<summary>Example: call a configured external API</summary>

Use this pattern only with a configured API connector.

```
Call the call_api function:
- path: 1.0/interactions/{customer_id}/save_summary
- api_code: crm_interactions
- method: POST
- payload: {"interaction_summary": "{summary}"}
```

</details>

<details>

<summary>Example: validate data with JSON logic</summary>

```
Call validate function with the following logic and data:
- logic: {"==": [{"var": "result"}, "200 OK"]}
- data: {"result": <call_api response>}
```

</details>

### Presentation

Use **Presentation** to configure how the assistant's responses appear in the Deepdesk widget.

| Field            | Description                                                                                  |
| ---------------- | -------------------------------------------------------------------------------------------- |
| **Display Name** | Name shown in the widget header. Leave this field empty to use the assistant's default name. |
| **Editable**     | Allows users to edit the assistant's response directly in the widget before using it.        |

### Knowledge

Use **Knowledge** to attach a knowledge source to the assistant when this option is available in the environment. Use **Select a base** to connect the assistant to a Knowledge Base or decision tree.

Attach a Knowledge Base when the assistant must answer from controlled source material instead of relying only on the transcript and prompt. Knowledge Bases are managed from **Admin Settings** > **Knowledge Base**. See [Manage Dialogue AI Assist Knowledge Bases](/channels-integrations/dialogue-ai-assist/manage-dialogue-ai-assist-knowledge-bases.md).

### Settings

Use **Settings** to configure runtime options for the assistant.

| Setting                 | Description                                                                                                       |
| ----------------------- | ----------------------------------------------------------------------------------------------------------------- |
| **Model**               | Selects the AI model used for this assistant. This field is required.                                             |
| **Include transcript**  | Adds the voice transcript or chat conversation to the prompt.                                                     |
| **Run Silently**        | Executes the assistant in the background without visible messages.                                                |
| **Reuse Thread**        | Includes previous interactions with this assistant for context.                                                   |
| **Error Handling Mode** | Controls how the assistant handles errors during execution. Options are **Retry on error** and **Stop on error**. |

The **Settings** tab also contains a **Delete** button for removing the assistant.

## Configure integration

Use the **Integration** tab to control whether another assistant can call this assistant as a tool.

Enable **Allow other assistants to call \[assistant name] as tool** to make the assistant available as a tool. When this setting is enabled, configure:

* **Tool Description**: Describe what the assistant does and when another assistant should call it.
* **Properties**: Define the parameters that another assistant can pass when it calls this assistant.

**Properties** has **Form** and **JSON** modes. In **Form** mode, define input fields, types, and descriptions using JSON Schema. Required fields determine what must be provided. Supported types include **String**, **Integer**, **Number**, and **Boolean**.

In **JSON** mode, enter the JSON schema directly. Use **Prettify JSON** to format the schema.

## Configure run control

Use the **Run Control** tab to configure the conditions and permissions that control whether the assistant runs.

| Field                  | Description                                                                                                                                                                |
| ---------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Allowed Groups**     | Restricts assistant availability to specific groups. Leave this field empty to allow all groups. The dropdown uses groups from **Agentic AI** > **Settings** > **Groups**. |
| **Run Condition**      | Defines the JSON logic that must evaluate to true before the assistant runs. If it evaluates to false, the assistant is skipped.                                           |
| **Fallback Assistant** | Defines the assistant to use as a backup when this assistant fails or cannot respond. The dropdown uses assistants from **Agentic AI** > **Assistants**.                   |

Use **Triggers** in **Run Control** to review where the assistant is used:

* **Event Based Assistant Assignments**
* **On-Demand Assistant Assignments**
* **Linked Assistants**

When no trigger data is available, the section shows **No triggers found**.

## Configure shared assistant settings

Use **Agentic AI** > **Settings** to configure shared assistant resources.

### Connectors

Assistant Connectors define APIs that can become tools for assistants. The overview shows connector cards with the connector code, name, and base URL. Select **Add API** to create an API connector.

In **Tools**, use **Generate tools from OpenAPI spec** to make API endpoints available as assistant tools. Provide a **URL to openapi.json** and select **Import**, or paste the OpenAPI specification into the text area. Use **Prettify JSON** and **Save** as needed.

In **Settings**, configure **Code \***, **URL \***, **Headers**, and authentication. Authentication options are **None**, **OAuth2**, and **Auth Headers**.

Microsoft Copilot Studio connectors make Copilot-backed tools available to assistants. Select **Add Copilot Agent** to create a connector.

### Routes

Use the **Routes** tab to manage assistant routes. Select **+ Add assistant route** to open the drawer. The route table contains **Assistant Code**, **Assistant Name**, and **Tags**.

### Groups

Use the **Groups** tab to define groups that can be selected in **Run Control** > **Allowed Groups**. The tab contains **Add Group**, search, and a table with **Name** and **In use for**.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://atlas.anywhere.now/channels-integrations/dialogue-ai-assist/manage-dialogue-ai-assist-assistants.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.