Grafana Alloy is the new name for our distribution of the OTel collector. Grafana Agent has been deprecated and is in Long-Term Support (LTS) through October 31, 2025. Grafana Agent will reach an End-of-Life (EOL) on November 1, 2025. Read more about why we recommend migrating to Grafana Alloy.
DocumentationGrafana AgentFlow modeTutorialsFlow by exampleFirst components and introducing the standard library
Open source

First components and the standard library

This tutorial covers the basics of the River language and the standard library. It introduces a basic pipeline that collects metrics from the host and sends them to Prometheus.

River basics

Recommended reading

River is an HCL-inspired configuration language used to configure Grafana Agent Flow. A River file is comprised of three things:

  1. Attributes

    key = value pairs used to configure individual settings.

    url = "http://localhost:9090"
  2. Expressions

    Expressions are used to compute values. They can be constant values (for example, "localhost:9090"), or they can be more complex (for example, referencing a component’s export: prometheus.exporter.unix.targets. They can also be a mathematical expression: (1 + 2) * 3, or a standard library function call: env("HOME")). We will use more expressions as we go along the examples. If you are curious, you can find a list of available standard library functions in the Standard library documentation.

  3. Blocks

    Blocks are used to configure components with groups of attributes or nested blocks. The following example block can be used to configure the logging output of Grafana Agent Flow:

    logging {
        level  = "debug"
        format = "json"


    The default log level is info and the default log format is logfmt.

    Try pasting this into config.river and running /path/to/agent run config.river to see what happens.

    Congratulations, you’ve just written your first River file! You’ve also just written your first Grafana Agent Flow configuration file. This configuration won’t do anything, so let’s add some components to it.


    Comments in River are prefixed with // and are single-line only. For example: // This is a comment.


Recommended reading

Components are the building blocks of a Grafana Agent Flow configuration. They are configured and linked to create pipelines that collect, process, and output your telemetry data. Components are configured with Arguments and have Exports that may be referenced by other components.

Let’s look at a simple example pipeline:

local.file "example" {
    path = env("HOME") + "file.txt"

prometheus.remote_write "local_prom" {
    endpoint {
        url = "http://localhost:9090/api/v1/write"

        basic_auth {
            username = "admin"
            password = local.file.example.content


A list of all available components can be found in the Component reference. Each component has a link to its documentation, which contains a description of what the component does, its arguments, its exports, and examples.

This pipeline has two components: local.file and prometheus.remote_write. The local.file component is configured with a single argument, path, which is set by calling the env standard library function to retrieve the value of the HOME environment variable and concatenating it with the string "file.txt". The local.file component has a single export, content, which contains the contents of the file.

The prometheus.remote_write component is configured with an endpoint block, containing the url attribute and a basic_auth block. The url attribute is set to the URL of the Prometheus remote write endpoint. The basic_auth block contains the username and password attributes, which are set to the string "admin" and the content export of the local.file component, respectively. The content export is referenced by using the syntax local.file.example.content, where local.file.example is the fully qualified name of the component (the component’s type + its label) and content is the name of the export.

Flow of example pipeline with local.file and prometheus.remote_write components


The local.file component’s label is set to "example", so the fully qualified name of the component is local.file.example. The prometheus.remote_write component’s label is set to "local_prom", so the fully qualified name of the component is prometheus.remote_write.local_prom.

This example pipeline still doesn’t do anything, so let’s add some more components to it.

Shipping your first metrics

Recommended reading

Make a simple pipeline with a prometheus.exporter.unix component, a prometheus.scrape component to scrape it, and a prometheus.remote_write component to send the scraped metrics to Prometheus.

prometheus.exporter.unix "localhost" {
    // This component exposes a lot of metrics by default, so we will keep all of the default arguments.

prometheus.scrape "default" {
    // Setting the scrape interval lower to make it faster to be able to see the metrics
    scrape_interval = "10s"

    targets    = prometheus.exporter.unix.localhost.targets
    forward_to = [

prometheus.remote_write "local_prom" {
    endpoint {
        url = "http://localhost:9090/api/v1/write"

Run Grafana Agent Flow with:

/path/to/agent run config.river

Navigate to http://localhost:3000/explore in your browser. After ~15-20 seconds, you should be able to see the metrics from the prometheus.exporter.unix component! Try querying for node_memory_Active_bytes to see the active memory of your host.

Screenshot of node_memory_Active_bytes query in Grafana

Visualizing the relationship between components

The following diagram is an example pipeline:

Flow of example pipeline with a prometheus.scrape, prometheus.exporter.unix, and prometheus.remote_write components

The preceding configuration defines three components:

  • prometheus.scrape - A component that scrapes metrics from components that export targets.
  • prometheus.exporter.unix - A component that exports metrics from the host, built around node_exporter.
  • prometheus.remote_write - A component that sends metrics to a Prometheus remote-write compatible endpoint.

The prometheus.scrape component references the prometheus.exporter.unix component’s targets export, which is a list of scrape targets. The prometheus.scrape component then forwards the scraped metrics to the prometheus.remote_write component.

One rule is that components can’t form a cycle. This means that a component can’t reference itself directly or indirectly. This is to prevent infinite loops from forming in the pipeline.

Exercise for the reader

Recommended Reading

Let’s start a container running Redis and configure Grafana Agent Flow to scrape metrics from it.

docker container run -d --name flow-redis -p 6379:6379 --rm redis

Try modifying the pipeline to scrape metrics from the Redis exporter. You can refer to the prometheus.exporter.redis component documentation for more information on how to configure it.

To give a visual hint, you want to create a pipeline that looks like this:

Flow of exercise pipeline, with a scrape, unix_exporter, redis_exporter, and remote_write component


You may find the concat standard library function useful.

You can run Grafana Agent Flow with the new configuration file by running:

/path/to/agent run config.river

Navigate to http://localhost:3000/explore in your browser. After the first scrape, you should be able to query for redis metrics as well as node metrics.

To shut down the Redis container, run:

docker container stop flow-redis

If you get stuck, you can always view a solution here:

Finishing up and next steps

You might have noticed that running Grafana Agent Flow with the configurations created a directory called data-agent in the directory you ran Grafana Agent Flow from. This directory is where components can store data, such as the prometheus.exporter.unix component storing its WAL (Write Ahead Log). If you look in the directory, do you notice anything interesting? The directory for each component is the fully qualified name.

If you’d like to store the data elsewhere, you can specify a different directory by supplying the --storage.path flag to Grafana Agent’s run command, for example, /path/to/agent run config.river --storage.path /etc/grafana-agent. Generally, you can use a persistent directory for this, as some components may use the data stored in this directory to perform their function.

In the next tutorial, you will look at how to configure Grafana Agent Flow to collect logs from a file and send them to Loki. You will also look at using different components to process metrics and logs before sending them.