
1. How collection has evolved at Grafana #
To understand why there are several agents in the Grafana ecosystem, you need to look at the history of observability.
From fragmentation to unification #
Available in: ✓ OSS ✓ Enterprise ✓ Cloud

Historically, observability was very siloed: Prometheus for metrics, Loki for logs and Tempo for traces. Each of these backends needed its own collection agent. With the adoption of unified observability (the LGTM stack) and the emergence of the OpenTelemetry standard, Grafana Labs changed its strategy, moving from multiple specialized agents to a single universal collector.
2. Promtail: the logs pioneer #
The historical agent dedicated to Grafana Loki, at end of life since March 2, 2026.
Simple and effective #
Promtail was created at the same time as Loki. Its role is very specific: discover log files (often on Kubernetes nodes or Linux servers), attach labels to them in the Prometheus way, and ship them to Loki.
- Strengths: very lightweight, simple YAML configuration, perfect if you only need to collect logs.
- Weaknesses: does not handle metrics or traces. If you want metrics, you must install Prometheus alongside it.
3. Grafana Agent: the first attempt at unification #
The multi-signal agent, now end-of-life.
The Swiss Army knife in YAML #
Faced with the proliferation of agents, Grafana Agent was launched to bring together in a single binary the features of Prometheus (scraping), Promtail (logs) and a trace receiver (Tempo/Jaeger/Zipkin).
- Strengths: a single binary to deploy to collect the 3 pillars of observability. Ready-to-use exporters built in (MySQL, NGINX, etc.).
- Weaknesses: its YAML configuration quickly became monolithic and hard to maintain on large infrastructures. Moreover, Grafana Agent reached end of life on November 1, 2025 (no more bug fixes or security updates) and has been replaced by Alloy.
4. Grafana Alloy: the new generation #
The natural successor, based on OpenTelemetry and configuration as code.
Programmable pipelines and native OTLP #
Grafana Alloy officially replaces Grafana Agent. It drops rigid YAML for a declarative configuration language called River, now named the “Alloy configuration syntax” (similar to Terraform's HCL). It is in fact an extended distribution of the OpenTelemetry Collector.
- Strengths: ultra-flexible configuration (conditional routing, in-flight data transformation), a built-in local web UI to debug pipelines (as a DAG graph), hot reload, and 100% native OpenTelemetry (OTLP) support in addition to Prometheus and Loki.
- Weaknesses: requires learning the Alloy syntax, which introduces a slight initial learning curve.
5. Head-to-head comparison and recommendations #
Which one should you install or keep today?
Summary table #
| Feature | Promtail | Grafana Agent | Grafana Alloy |
|---|---|---|---|
| Project status | End of life (March 2, 2026) | End of life (November 1, 2025) | Recommended (active) |
| Signals collected | Logs only | Logs, metrics, traces | Logs, metrics, traces, profiles |
| Configuration format | YAML | Static YAML or Flow | Alloy syntax (formerly River) (programmable) |
| OpenTelemetry (OTLP) support | No | Partial | Native and complete |
| Local debugging UI | No | No | Yes (DAG visualization) |
The verdict #
Today the choice is clear: for any new deployment, use Grafana Alloy. It performs much better, is more flexible and represents the future of telemetry collection at Grafana. If you currently use Grafana Agent, it is strongly recommended to plan a migration to Alloy.
Finally, Promtail also reached end of life on March 2, 2026: it no longer receives fixes, including security fixes. It should no longer be chosen, even for a logs-only need: migrate with alloy convert --source-format=promtail, which translates the YAML configuration into Alloy configuration.
6. Go further with Alloy #
Put your knowledge into practice with our dedicated course.
Our recommendations #
If you want to master Grafana Alloy's new declarative pipelines, discover our complete course:
