Prompt
I'm building a cloud-native app and want vendor-neutral telemetry with OpenTelemetry, what should I pick?
Latest observation
Oct 19, 2026
OpenAI APIWeb search: off
For a vendor-neutral cloud-native telemetry setup, the best default choice is:
Pick OpenTelemetry + the OpenTelemetry Collector
Why this is the best foundation
- Vendor-neutral: you can send telemetry to any backend later.
- Cloud-native friendly: works well with microservices, Kubernetes, autoscaling, and distributed systems.
- Standardized data model: traces, metrics, and logs use common conventions.
- Portable: you can change observability vendors without rewriting your app instrumentation.
Recommended architecture
- Instrument your app with OpenTelemetry SDKs
- Use language-specific OTel libraries for traces, metrics, and logs.
- Send telemetry to the OpenTelemetry Collector
- Run it as a sidecar, DaemonSet, or central service depending on your environment.
- Export from the Collector to your chosen backend
- Examples: Grafana Tempo, Prometheus, Loki, Jaeger, Datadog, New Relic, Honeycomb, etc.
What to choose by signal
Traces
- OpenTelemetry tracing API/SDK
- Export using OTLP to the Collector
Metrics
- OpenTelemetry metrics API/SDK
- Prefer OTLP metrics end-to-end where possible
- In Kubernetes, you may still integrate with Prometheus style scraping depending on your stack
Logs
- Use OpenTelemetry logs where supported in your language/runtime
- In practice, many teams still route logs through standard log pipelines and enrich them with trace/span context
Practical default stack
If you want a clean vendor-neutral baseline:
- App instrumentation: OpenTelemetry SDKs
- Transport: OTLP
- Aggregation/processing: OpenTelemetry Collector
- Backends:
- Traces: Grafana Tempo or Jaeger
- Metrics: Prometheus or Grafana Mimir
- Logs: Grafana Loki
If you want the simplest starting point
Use:
- OpenTelemetry auto-instrumentation for your language
- OTLP exporter
- OpenTelemetry Collector
- One backend of your choice
Rule of thumb
If your goal is vendor neutrality, avoid tying your app directly to a proprietary agent or SDK.
Instrument once with OpenTelemetry, export through the Collector, choose the backend later.
If you want, I can also recommend a concrete setup for Kubernetes, AWS, Azure, GCP, or a specific language like Java, Go, Node.js, Python, or .NET.