> For the complete documentation index, see [llms.txt](https://documentation.proto.cx/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://documentation.proto.cx/docs/modules/inbox/inbox-co-pilot.md).

# Inbox co-pilot

{% hint style="info" %}
Inbox co-pilot is fully powered by Proto-hosted LLM models. Data is not exchanged with third parties. Questions and responses remain private and are never visible to the person.
{% endhint %}

{% hint style="warning" %}
Inbox co-pilot is available on all plans. Using it requires at least the **View livechats assigned to the user** permission. Configuring it requires the **Manage AI agents** permission.
{% endhint %}

Inbox co-pilot is a Proto-hosted LLM assistant available for livechats and tickets. It can use conversation context, person history, and a dedicated knowledge base to answer questions and recommend responses.

***

## Ask the co-pilot

1. Go to **Inbox**.
2. Select a livechat or ticket.
3. Select **Co-pilot** in the conversation details panel.
4. Enter a question.

Use the co-pilot to draft a reply, summarise the conversation, or retrieve information from its knowledge base.

***

## Draft a reply

1. Select a conversation in the Inbox.
2. Select **Draft reply** above the message composer.

The Co-pilot tab opens automatically and asks, **What should I reply?** The co-pilot uses the current conversation to suggest a response.

Select one of the following:

| Option | Result                                                                                         |
| ------ | ---------------------------------------------------------------------------------------------- |
| Accept | Places the suggested reply in the composer without sending it. You can edit it before sending. |
| Deny   | Dismisses the suggestion and leaves the composer unchanged.                                    |

***

## Mediate a conversation

Mediate mode uses more extensive reasoning to help the live agent understand and resolve a case. Its response can analyse the situation, possible root cause, customer sentiment and stakes, available options, recommended resolution, and next steps or risks.

1. Select **Mediate** above the message composer.
2. Ask a question in the Co-pilot tab.

Mediate mode remains active across subsequent questions. Select **Mediate** again to turn it off.

If you ask for a suggested reply, the co-pilot provides **Accept** and **Deny** controls. Accepting the suggestion places it in the composer without sending it.

{% hint style="warning" %}
Mediation requires access to **the current conversation**.
{% endhint %}

***

## Configure Inbox co-pilot

Go to **Settings → Teamspace → Inbox co-pilot**. Configuration is split across the **Agent** and **Knowledge** tabs.

### Agent

**Data access**

Choose which information the co-pilot can use when answering live agents.

| Setting                                                  | Description                                                                                   |
| -------------------------------------------------------- | --------------------------------------------------------------------------------------------- |
| Access all conversations                                 | Allows the co-pilot to reference any conversation in the workspace.                           |
| Access current person's profile and conversation history | Allows the co-pilot to reference the open person's profile and previous conversations.        |
| Access current conversation                              | Allows the co-pilot to reference and mediate the active conversation. Required for mediation. |

***

#### **Personalisation**

Select the LLM used by the co-pilot and edit its personality prompt. The prompt defines the co-pilot's role, environment, tone, goals, and general behaviour.

Select **Save** to confirm your changes.

***

### Knowledge

The Inbox co-pilot has a dedicated knowledge base. These sources provide information to the live-agent assistant and are separate from the sources used to train customer-facing AI agents.

Select one of the three options at the top of the **Knowledge base** table.

**Upload a file**

Supported file types: **CSV, JSON, PDF**\
Maximum file size: **5 MB**

1. Select **Upload file**.
2. Enter a name for the source.
3. Drag and drop the file or select **Browse Files**.
4. Select **Create**.

***

**Add a URL**

1. Select **Add URL**.
2. Enter a name and the full HTTPS URL.
3. Choose a scope:
   * **URL webpage** uses only the page at the specified URL.
   * **All webpages** crawls and trains on pages linked from the URL.
4. Optionally enable automatic retraining and select an interval.
5. Select **Create**.

***

**Paste a snippet**

1. Select **Paste snippet**.
2. Enter a name.
3. Paste or enter the content in **Text snippet**.
4. Select **Create**.

***

**Manage knowledge sources**

The Knowledge base table displays each source's name, last-updated date, and status.

| Status  | Meaning                                         |
| ------- | ----------------------------------------------- |
| Pending | Queued for training                             |
| Loading | Training in progress                            |
| Success | Trained and active                              |
| Fail    | Training failed—remove the source and try again |

To delete a source, select its row and use the delete option.

***

## Related

{% content-ref url="/pages/uv13kzt9HmAPCK2fDyRc" %}
[Inbox](/docs/modules/inbox.md)
{% endcontent-ref %}

{% content-ref url="/pages/rvah6ZvOyA8ja414gax6" %}
[Message editor](/docs/modules/inbox/message-editor.md)
{% endcontent-ref %}

{% content-ref url="/pages/iYbvOfviA4QyuUMR2E7a" %}
[Roles & permissions](/docs/settings/general/roles-and-permissions.md)
{% endcontent-ref %}

{% content-ref url="/pages/XXSVxZ3ZK1tPNmNWlOxu" %}
[Knowledge](/docs/modules/ai-agents/knowledge.md)
{% endcontent-ref %}