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

# Datadog AI Guard

> Connect Datadog AI Guard to inference hooks, manage credentials, and investigate evaluations in Datadog.

Connect [Datadog AI Guard](https://docs.datadoghq.com/security/ai_guard/) as a provider for Dust inference hooks. This guide covers Datadog access, connection settings, credentials, and investigation in Datadog. For checkpoints, evaluation context, enforcement, and workspace-wide controls, see the [Inference hooks overview](https://docs.dust.tt/docs/user-documentation/admins/admin-governance/inference-hooks.md).

> [!NOTE]
>
> **Private beta.** Contact your Dust team or [inference-hooks@dust.com](mailto:inference-hooks@dust.com) to request access and arrange onboarding.

## Before you start

You need:

* Inference hooks enabled for your Dust workspace.
* Access to Datadog AI Guard for your Datadog organization.
* A Datadog API key and an application key with the `ai_guard_evaluate` scope.

Confirm your Datadog access and commercial terms with your Datadog team before enabling the integration. Datadog usage is subject to your agreement with Datadog.

You must be a workspace admin to create or edit hooks. Enabling this integration sends the selected content to Datadog for evaluation, including when the hook is in monitor mode.

## Connect Datadog AI Guard

1. Go to **Admin > Security > Inference Hooks**.
2. Click **Add your first hook**, or **Add hook** if you already have one.
3. Enter a descriptive **Name**.
4. Select **Datadog AI Guard** as the **Provider**.
5. Select your **Datadog site**, then enter your **API key** and **Application key**.
6. Click **Test connection**. This checks the keys and access to AI Guard without evaluating conversation content. A successful test does not validate your security policies.
7. Set **Service** and **Environment** to identify the integration in Datadog. The defaults are `dust` and `production`.
8. Keep **Enforcement** set to **Monitor** for your initial rollout.
9. Select one or more **Events**, then choose how much conversation history to include under **Evaluation context**. See [Choose evaluation checkpoints](https://docs.dust.tt/docs/user-documentation/admins/admin-governance/inference-hooks.md#choose-evaluation-checkpoints) and [Choose evaluation context](https://docs.dust.tt/docs/user-documentation/admins/admin-governance/inference-hooks.md#choose-evaluation-context) in the overview.
10. Review **Excluded groups**, **When the hook fails or times out**, and **Timeout (ms)**, then click **Create**.

Creating the hook activates it across your workspace, subject to group exclusions. Monitor mode does not limit the data sent to Datadog: review the data-sharing and billing implications before creating the hook, then use monitor mode to assess findings from actual agent activity before turning on blocking.

## Investigate in Datadog

Dust sends an evaluation trace to Datadog after receiving an AI Guard verdict. In Dust, open **Admin > Security > Inference Hooks**, select an evaluation, and use **View in Datadog** when available to investigate there. You can also use the provider reference to correlate the records.

Trace delivery is separate from the verdict: an evaluation can be recorded in Dust even if its trace is not available in Datadog. The shared [Review evaluations](https://docs.dust.tt/docs/user-documentation/admins/admin-governance/inference-hooks.md#review-evaluations) section explains Dust outcomes and how to investigate a user's Dust reference.

## Update credentials

Edit the hook to update its Datadog connection settings. Leave both key fields empty to keep the current keys. Changing the Datadog site requires new keys.

## Troubleshoot the connection

If **Test connection** rejects the keys, confirm that both keys belong to the selected Datadog site, that the application key has `ai_guard_evaluate`, and that your organization can use AI Guard.

If evaluations report a provider error, check their error details for invalid credentials, rate limits, or timeouts. Review the hook's failure behavior before changing its timeout. For other issues, see [Troubleshooting](https://docs.dust.tt/docs/user-documentation/admins/admin-governance/inference-hooks.md#troubleshooting) in the overview.

Share Datadog integration issues with your Dust team or [inference-hooks@dust.com](mailto:inference-hooks@dust.com).


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.