> Documentation index for AI agents: https://docs.pagerduty.com/llms.txt

# Prometheus Self-Hosted

Connect a Prometheus instance inside your network through the PD Automation Runner and run PromQL queries from Incident Workflows.

{/* Hidden from navigation until PD Automation Runner GA. */}

> **Early Access:** This feature is currently in Early Access (EA). Click [here](https://www.pagerduty.com/early-access/) to sign up for the Early Access program. This feature is only available for PD Reliability Platform accounts.

Use the Prometheus Self-hosted connection for a Prometheus instance inside your network. PagerDuty sends requests through a [PD Automation Runner](/ai-automation/automation/pd-automation-runner) running in your environment, so responders can query Prometheus from an Incident Workflow without exposing it to the internet.

## Requirements

- A PD Automation Runner connection whose runner can reach your Prometheus instance. To set one up, see [PD Automation Runner](/ai-automation/automation/pd-automation-runner). The runner can use any deployment method, and a single runner can serve Kubernetes, Prometheus, and Grafana connections.

## Add a Prometheus Self-Hosted Connection

1. In PagerDuty, navigate to **Integrations**  →  **Automation Connectors**.
2. Select **Prometheus Self-hosted** and click **Add Connector**.
3. Enter the following fields:

| Field | Description |
| :--- | :--- |
| **Connection Name** | The name of the new connection. |
| **Runner Connection** | The PD Automation Runner connection that can reach this Prometheus instance. |
| **Prometheus URL** | The base URL of your Prometheus instance as reachable from the runner, for example `http://prometheus.monitoring.svc.cluster.local:9090`. Do not include `/api`. |
| **Authentication** | The authentication the runner uses: **None**, **Basic**, or **Bearer**. Defaults to **None**. Prometheus has no built-in authentication, so choose **Basic** or **Bearer** only if a reverse proxy or similar service sits in front of it. |
| **Username** | The username for Basic authentication. Ignored for None and Bearer. |
| **Password** | The Basic authentication password, or the Bearer token. Ignored for None. |

4. Click **Save**.

### How PagerDuty Validates the Connection

When you save the connection, the runner calls `GET {Prometheus URL}/api/v1/status/buildinfo` with the configured authentication. PagerDuty creates the connection only if Prometheus responds successfully. Editing the runner, URL, or authentication runs the same check. Editing only the name does not.

### Troubleshoot the Connection

| Error | What to Check |
| :--- | :--- |
| `Runner failed to reach Prometheus: …` | The runner cannot reach the URL. Check the host, port, and the runner's network access. |
| `Prometheus returned a non-JSON response…` | The URL responded, but it is not the Prometheus API. For example, it is a login page, another service, or a URL that already includes `/api`. |
| `Prometheus rejected the request as "Unauthorized"…` | A proxy in front of Prometheus rejected the credentials. Check **Authentication**, **Username**, and **Password**. |
| `Prometheus rejected the connection check: …` | Prometheus returned an error. The message includes the reason from Prometheus. |

## Prometheus Workflow Actions

These actions require a Prometheus Self-hosted connection routed through a PD Automation Runner.

To add an action to a workflow:

1. If you have not done so, [create an Incident Workflow](/ai-automation/automation/incident-workflows).
2. When the instructions prompt you to add actions, select the action.
3. Enter the inputs and click **Save**.
4. Publish the workflow.

When the action runs, it returns the outputs listed for that action.

> **Field References:** Fields with the  icon accept Field References, which reference incident data or outputs from prior workflow steps. To add a Field Reference, click  or enter `{{`, then select the field.

### Instant Query

Evaluates a PromQL expression at a single point in time. The query runs on your runner inside your network, so responders can check the current state of a service, such as whether its targets are up or its current error rate, without leaving the workflow.

| Input | Description |
| :--- | :--- |
| **Connection Input** | The Prometheus Self-hosted connection to use. |
| **Query** | The PromQL expression to evaluate. For example: `up`, `rate(http_requests_total[5m])`, or `sum by (pod) (container_memory_usage_bytes)`. |
| **Time** (Optional) | The evaluation timestamp, as RFC3339 (for example, `2026-10-06T14:00:00Z`) or Unix seconds. If left blank, Prometheus evaluates the query at its current server time. |
| **Include Query Statistics** (Optional) | Also returns Prometheus timing and sample statistics. Defaults to off. |

| Output | Description |
| :--- | :--- |
| **Results** | An array of series, each with its labels and an array of `{timestamp, value, valueRaw}` data points. |
| **Result Type** | The Prometheus result type: `vector`, `scalar`, `string`, or `matrix`. |
| **Series Count** | The number of series returned. |
| **Statistics** | Prometheus timing and sample statistics. Empty unless **Include Query Statistics** is enabled. |
| **Result** | Whether the action succeeded: `Success` or `Failed`. |
| **Result Summary** | A brief description of what the action did, or why it failed. |
| **Error** | A brief description, populated if the action failed. |

> **Keep Queries Focused:** Use label matchers and aggregation, for example `sum by (pod) (...)` or `{job="api"}`, to return only the series responders need. Broad queries can return a large number of series and make the output harder to act on.

### Range Query

Evaluates a PromQL expression over a time range. The query runs on your runner inside your network, so responders can see how a metric changed leading up to an incident, such as error rate or latency over the last hour, without leaving the workflow.

| Input | Description |
| :--- | :--- |
| **Connection Input** | The Prometheus Self-hosted connection to use. |
| **Query** | The PromQL expression to evaluate. For example: `up`, `rate(http_requests_total[5m])`, or `sum by (pod) (container_memory_usage_bytes)`. |
| **Start** | The start of the time range, as RFC3339 (for example, `2026-10-06T13:00:00Z`) or Unix seconds. |
| **End** | The end of the time range, as RFC3339 or Unix seconds. Must not be before **Start**. |
| **Step** (Optional) | The resolution between data points, for example `15s`, `1m`, or `5m`. Defaults to `1m`. |
| **Include Query Statistics** (Optional) | Also returns Prometheus timing and sample statistics. Defaults to off. |

| Output | Description |
| :--- | :--- |
| **Results** | An array of series, each with its labels and an array of `{timestamp, value, valueRaw}` data points. |
| **Result Type** | The Prometheus result type. `matrix` for a successful range query. |
| **Series Count** | The number of series returned. |
| **Total Data Points** | The total number of data points across all series. |
| **Statistics** | Prometheus timing and sample statistics. Empty unless **Include Query Statistics** is enabled. |
| **Result** | Whether the action succeeded: `Success` or `Failed`. |
| **Result Summary** | A brief description of what the action did, or why it failed. |
| **Error** | A brief description, populated if the action failed. |

> **Choose a Step:** Prometheus rejects a range query that returns more than 11,000 data points per series. For long ranges, use a larger **Step**. For example, a 24-hour range needs a step of at least `8s`, and `1m` or `5m` is usually enough for incident triage.

## Learn More

- [PD Automation Runner](/ai-automation/automation/pd-automation-runner)
- [PD Automation Runner: Installation](/ai-automation/automation/pd-automation-runner/pd-runner-installation)
- [Incident Workflows](/ai-automation/automation/incident-workflows)