diff --git a/src/content/docs/guides/analytics/collect-metrics.mdx b/src/content/docs/guides/analytics/collect-metrics.mdx
index 24bf654f0..a24e94b95 100644
--- a/src/content/docs/guides/analytics/collect-metrics.mdx
+++ b/src/content/docs/guides/analytics/collect-metrics.mdx
@@ -31,7 +31,7 @@ the snapshot of all historical metrics.
 ## Summarize metrics
 
 You can [transform](/guides/transformation/filter-and-select-data) metrics like
-ordinary data, e.g., write aggregations over metrics to compute runtime
+ordinary data, for example, write aggregations over metrics to compute runtime
 statistics suitable for reporting or dashboarding:
 
 ```tql
@@ -41,8 +41,8 @@ summarize runtime=sum(duration), pipeline_id
 sort -runtime
 ```
 
-The above example computes the total runtime over all pipelines grouped by their
-unique ID.
+The previous example computes the total runtime over all pipelines grouped by
+their unique ID.
 
 ## Inspect pipeline throughput
 
@@ -57,3 +57,16 @@ sort -egress_events
 
 The `events` fields count records that passed through the pipeline during the
 metric period. The `bytes` fields approximate the amount of data transferred.
+
+## Send metrics to Prometheus
+
+Use `shape="prometheus"` with <Op>metrics</Op> and <Op>to_prometheus</Op> to
+send live metrics to a Prometheus-compatible Remote Write receiver:
+
+```tql
+metrics live=true, shape="prometheus"
+to_prometheus "https://prometheus.example/api/v1/write"
+```
+
+The <Integration>prometheus</Integration> integration page covers endpoint
+setup, authentication headers, and additional examples.
diff --git a/src/content/docs/integrations/index.mdx b/src/content/docs/integrations/index.mdx
index 9e3899017..e3ac55493 100644
--- a/src/content/docs/integrations/index.mdx
+++ b/src/content/docs/integrations/index.mdx
@@ -33,8 +33,9 @@ communication over numerous protocols and APIs:
 - **Message queues**: <Integration>kafka</Integration>,
   <Integration>nats</Integration>, <Integration>amazon/sqs</Integration>,
   <Integration>amqp</Integration>
-- **Databases**: <Integration>snowflake</Integration>,
-  <Integration>clickhouse</Integration>, <Integration>mysql</Integration>
+- **Databases and analytics**: <Integration>snowflake</Integration>,
+  <Integration>clickhouse</Integration>, <Integration>mysql</Integration>,
+  <Integration>prometheus</Integration>
 - **Network protocols**: <Integration>tcp</Integration>,
   <Integration>udp</Integration>, <Integration>http</Integration>,
   <Integration>syslog</Integration>
diff --git a/src/content/docs/integrations/prometheus.mdx b/src/content/docs/integrations/prometheus.mdx
new file mode 100644
index 000000000..0da0bb290
--- /dev/null
+++ b/src/content/docs/integrations/prometheus.mdx
@@ -0,0 +1,67 @@
+---
+title: Prometheus
+---
+
+[Prometheus](https://prometheus.io/) is an open-source monitoring system and
+time-series database. Tenzir can send metric events to Prometheus-compatible
+Remote Write receivers, including Prometheus, Grafana Mimir, Cortex, Thanos
+Receive, and VictoriaMetrics.
+
+Use <Op>metrics</Op> with `shape="prometheus"` and <Op>to_prometheus</Op> to
+send Tenzir node metrics to a Prometheus Remote Write endpoint. This integration
+doesn't expose a scrape endpoint for the Prometheus pull model.
+
+## Prerequisites
+
+Before you send metrics, configure your receiver to accept Remote Write traffic.
+For Prometheus, start the server with the Remote Write receiver enabled and use
+the `/api/v1/write` endpoint.
+
+Store endpoint URLs and authentication headers as
+[secrets](/explanations/secrets) when they contain credentials.
+
+## Examples
+
+### Send canonical metric events
+
+Send events that already use the default metric shape expected by
+<Op>to_prometheus</Op>:
+
+```tql
+from {
+  metric: "http_requests_total",
+  value: 42,
+  timestamp: 2026-05-15T10:00:00Z,
+  labels: {
+    method: "GET",
+    status: 200,
+  },
+}
+to_prometheus "https://prometheus.example/api/v1/write"
+```
+
+### Export Tenzir node metrics
+
+Convert live Tenzir node metrics to Prometheus samples:
+
+```tql
+metrics live=true, shape="prometheus"
+to_prometheus "https://prometheus.example/api/v1/write"
+```
+
+### Add authentication headers
+
+Pass additional HTTP headers when your receiver requires authentication. Header
+values are resolved as secrets.
+
+```tql
+metrics live=true, shape="prometheus"
+to_prometheus "prometheus-remote-write-url",
+  headers={Authorization: "prometheus-remote-write-token"}
+```
+
+## See Also
+
+- <Op>metrics</Op>
+- <Op>to_prometheus</Op>
+- <Guide>analytics/collect-metrics</Guide>
diff --git a/src/content/docs/reference/operators.mdx b/src/content/docs/reference/operators.mdx
index 22af14ac1..a65c5938b 100644
--- a/src/content/docs/reference/operators.mdx
+++ b/src/content/docs/reference/operators.mdx
@@ -723,6 +723,10 @@ operators:
     description: 'Sends events to an OpenSearch-compatible Bulk API.'
     example: 'to_opensearch "localhost:9200", …'
     path: 'reference/operators/to_opensearch'
+  - name: 'to_prometheus'
+    description: 'Sends metric events to a Prometheus Remote Write receiver.'
+    example: 'to_prometheus "https://prometheus.example/api/v1/write"'
+    path: 'reference/operators/to_prometheus'
   - name: 'to_s3'
     description: 'Writes events to one or multiple objects in Amazon S3.'
     example: 'to_s3 "s3://my-bucket/data/{uuid}.json" { write_ndjson }'
@@ -2050,6 +2054,14 @@ to_opensearch "localhost:9200", …
 
 </ReferenceCard>
 
+<ReferenceCard title="to_prometheus" description="Sends metric events to a Prometheus Remote Write receiver." href="/reference/operators/to_prometheus">
+
+```tql
+to_prometheus "https://prometheus.example/api/v1/write"
+```
+
+</ReferenceCard>
+
 <ReferenceCard title="to_s3" description="Writes events to one or multiple objects in Amazon S3." href="/reference/operators/to_s3">
 
 ```tql
diff --git a/src/content/docs/reference/operators/metrics.mdx b/src/content/docs/reference/operators/metrics.mdx
index 8211c2518..d6b779d2c 100644
--- a/src/content/docs/reference/operators/metrics.mdx
+++ b/src/content/docs/reference/operators/metrics.mdx
@@ -69,7 +69,8 @@ Controls the output shape. The default is `"raw"`, which preserves the native
 `tenzir.metrics.*` schemas listed below.
 
 Set `shape="prometheus"` to transform metrics into canonical records that are
-compatible with Prometheus-oriented pipelines:
+compatible with Prometheus-oriented pipelines. You can send this shape directly
+to <Op>to_prometheus</Op>.
 
 ```tql
 {
@@ -725,6 +726,13 @@ select timestamp, used_bytes
 {timestamp: 2023-12-21T12:55:32.916200, used_bytes: 461842751488}
 ```
 
+### Send metrics to Prometheus
+
+```tql
+metrics live=true, shape="prometheus"
+to_prometheus "https://prometheus.example/api/v1/write"
+```
+
 ### Get the memory usage over time
 
 ```tql
@@ -782,4 +790,7 @@ select timestamp, port, handle, reads, bytes
 ## See Also
 
 - <Op>diagnostics</Op>
+- <Op>to_prometheus</Op>
+- <Guide>analytics/collect-metrics</Guide>
 - <Tutorial>plot-data-with-charts</Tutorial>
+- <Integration>prometheus</Integration>
diff --git a/src/content/docs/reference/operators/to_prometheus.mdx b/src/content/docs/reference/operators/to_prometheus.mdx
new file mode 100644
index 000000000..227e18a27
--- /dev/null
+++ b/src/content/docs/reference/operators/to_prometheus.mdx
@@ -0,0 +1,277 @@
+---
+title: to_prometheus
+category: Outputs/Events
+example: 'to_prometheus "https://prometheus.example/api/v1/write"'
+---
+
+import Op from '@components/see-also/Op.astro';
+import TLSOptions from '@partials/operators/TLSOptions.mdx';
+
+Sends metric events to a Prometheus Remote Write receiver.
+
+```tql
+to_prometheus url:string, [protobuf_message=string, name=string,
+    value=number, timestamp=time, labels=record, type=string, help=string,
+    unit=string, family=string, start_timestamp=time, headers=record,
+    tls=record, timeout=duration, connection_timeout=duration,
+    max_retry_count=int, retry_delay=duration, flush_interval=duration,
+    max_samples_per_request=int, max_uncompressed_bytes=int,
+    sanitize_names=bool]
+```
+
+## Description
+
+The `to_prometheus` operator converts incoming events to Prometheus Remote
+Write requests and sends them with HTTP `POST`. The request body is a protobuf
+message compressed with Snappy.
+
+By default, the operator sends Remote Write v1 requests with the
+`prometheus.WriteRequest` protobuf message. Set
+`protobuf_message="io.prometheus.write.v2.Request"` to send Remote Write v2
+requests.
+
+The default input shape is:
+
+```tql
+{
+  metric: "http_requests_total",
+  value: 42,
+  timestamp: 2026-05-15T10:00:00Z,
+  labels: {
+    method: "GET",
+    status: 200,
+  },
+  type: "counter",
+  help: "HTTP requests",
+  unit: "requests",
+  family: "http_requests",
+  start_timestamp: 2026-05-15T09:00:00Z,
+}
+```
+
+You can use expression options such as `name=metric_name` or
+`labels={tenant: tenant_id}` to send existing schemas without reshaping them
+first.
+
+The <Op>metrics</Op> operator can produce the default input shape with
+`shape="prometheus"`.
+
+The operator writes the metric-name expression as the reserved Prometheus
+`__name__` label. The expression defaults to the `metric` field. Providing
+`labels.__name__` is invalid. Samples with the same label set are batched into
+one time series per request, and samples in each time series are sorted by
+timestamp before serialization.
+
+### `url: string`
+
+The Remote Write endpoint URL.
+
+The URL is resolved as a [secret](/explanations/secrets), so you can pass a
+secret name to avoid hardcoding sensitive URLs.
+
+### `protobuf_message = string (optional)`
+
+The protobuf message to send.
+
+Supported values:
+
+- `prometheus.WriteRequest`
+- `io.prometheus.write.v2.Request`
+
+Defaults to `prometheus.WriteRequest`.
+
+### `name = string (optional)`
+
+Expression for the metric name.
+
+Defaults to `metric`.
+
+### `value = number (optional)`
+
+Expression for the sample value.
+
+Defaults to `value`.
+
+### `timestamp = time (optional)`
+
+Expression for the sample timestamp.
+
+If the expression evaluates to `null` or the field is absent, Tenzir uses the
+processing time.
+
+Defaults to `timestamp`.
+
+### `labels = record (optional)`
+
+Expression for metric labels.
+
+The record field names become label names. Label values are converted to
+strings. Empty label names, empty label values, duplicate label names, and
+`__name__` labels are invalid.
+
+Defaults to `labels`. If the field is absent, the operator sends no custom
+labels.
+
+### `type = string (optional)`
+
+Expression for Prometheus metric metadata type.
+
+Supported values are `counter`, `gauge`, `histogram`, `gaugehistogram`,
+`summary`, `info`, and `stateset`.
+
+Defaults to `type`.
+
+### `help = string (optional)`
+
+Expression for the metric help text.
+
+Defaults to `help`.
+
+### `unit = string (optional)`
+
+Expression for the metric unit.
+
+Defaults to `unit`.
+
+### `family = string (optional)`
+
+Expression for the metric family name.
+
+If the expression evaluates to `null` or the field is absent, Tenzir uses the
+final metric name.
+
+Defaults to `family`.
+
+### `start_timestamp = time (optional)`
+
+Expression for the Remote Write v2 sample start timestamp.
+
+This field is only sent for Remote Write v2 requests. Defaults to
+`start_timestamp`.
+
+### `headers = record (optional)`
+
+Additional HTTP headers. Header values are resolved as
+[secrets](/explanations/secrets).
+
+The following Remote Write headers are reserved and cannot be overridden:
+
+- `Content-Encoding`
+- `Content-Length`
+- `Content-Type`
+- `User-Agent`
+- `X-Prometheus-Remote-Write-Version`
+
+### `timeout = duration (optional)`
+
+Timeout for each HTTP request.
+
+Defaults to `30s`.
+
+### `connection_timeout = duration (optional)`
+
+Timeout for establishing the connection.
+
+Defaults to `5s`.
+
+### `max_retry_count = int (optional)`
+
+Maximum number of retry attempts per request.
+
+A request is retried on transient transport failures and HTTP `429` and `5xx`
+responses.
+
+Defaults to `5`.
+
+### `retry_delay = duration (optional)`
+
+Base duration between retry attempts.
+
+Tenzir uses exponential backoff starting at `retry_delay` and capping at
+`16 * retry_delay`. A `Retry-After` response header overrides this delay.
+
+Defaults to `1s`.
+
+<TLSOptions tls_default="false"/>
+
+### `flush_interval = duration (optional)`
+
+Maximum amount of time to buffer samples before sending a request.
+
+Defaults to `1s`.
+
+### `max_samples_per_request = int (optional)`
+
+Maximum number of samples to include in one request.
+
+Defaults to `2000`.
+
+### `max_uncompressed_bytes = int (optional)`
+
+Maximum uncompressed protobuf request size in bytes.
+
+Defaults to `4Mi`.
+
+### `sanitize_names = bool (optional)`
+
+Whether to replace invalid metric-name and label-name characters with `_`.
+
+Defaults to `true`.
+
+## Examples
+
+### Send canonical metric events
+
+```tql
+from {
+  metric: "http_requests_total",
+  value: 42,
+  timestamp: 2026-05-15T10:00:00Z,
+  labels: {
+    method: "GET",
+    status: 200,
+  },
+}
+to_prometheus "https://prometheus.example/api/v1/write"
+```
+
+### Map an existing schema
+
+```tql
+from {
+  metric_name: "cpu_usage",
+  usage: 0.82,
+  host: "sensor-a",
+}
+to_prometheus "https://prometheus.example/api/v1/write",
+  name=metric_name,
+  value=usage,
+  labels={host: host}
+```
+
+### Send Remote Write v2 metadata
+
+```tql
+from {
+  metric: "requests_total",
+  value: 100,
+  type: "counter",
+  help: "Total requests",
+  unit: "requests",
+}
+to_prometheus "https://prometheus.example/api/v1/write",
+  protobuf_message="io.prometheus.write.v2.Request"
+```
+
+## Limitations
+
+The operator currently supports scalar samples only. It does not support native
+histograms, exemplars, persistent queues, automatic protocol negotiation,
+OAuth, SigV4, or round-robin DNS behavior.
+
+## See Also
+
+- <Op>metrics</Op>
+- <Op>to_http</Op>
+- <Guide>analytics/collect-metrics</Guide>
+- <Integration>prometheus</Integration>
diff --git a/src/sidebar.ts b/src/sidebar.ts
index ddbffd657..5b40dff8f 100644
--- a/src/sidebar.ts
+++ b/src/sidebar.ts
@@ -398,6 +398,7 @@ export const integrations = [
       "integrations/elasticsearch",
       "integrations/mysql",
       "integrations/opensearch",
+      "integrations/prometheus",
       "integrations/snowflake",
       "integrations/splunk",
     ],