Back to catalog
Odigos
Apache-2.0Odigos replaces manual OpenTelemetry instrumentation and proprietary runtime agents like OverOps with eBPF-driven, zero-code tracing for Kubernetes.
Open-source alternative to — compare it against:
Overview
Odigos is an observability control plane that provides distributed tracing and telemetry across cloud infrastructure without application source modifications. It pairs eBPF kernel instrumentation with native OpenTelemetry agents to detect and instrument microservices on the fly. Infrastructure teams orchestrate the entire telemetry lifecycle through declarative Kubernetes custom resources rather than maintaining per-service monitoring libraries.
Why Odigos instead of OverOps
OverOps relied on proprietary runtime agents injected into JVM and .NET startup flags to record in-memory variables during exceptions, requiring continuous maintenance of custom launch parameters across container images. Running that stack required specialized on-premises collector servers or routing diagnostic snapshots to a commercial platform. Odigos shifts operational overhead from application configuration to the Kubernetes control plane. An in-cluster operator and a node daemon named odiglet discover workloads, attach eBPF probes or inject standard OpenTelemetry agents dynamically, and manage auto-scaling collector gateways without modifying container manifests or build steps.
The trade-off centers on forensic depth versus architectural flexibility. OverOps delivered granular variable states at the exact line of failure, but locked teams into supported runtimes and closed tooling. Odigos outputs standard OTLP traces, metrics, and logs directly into backends such as Jaeger, Prometheus, Tempo, Datadog, or Sentry. Operations teams avoid maintaining bespoke agents for each language, but they must operate the OpenTelemetry collector layer and rely on external platforms for incident triage.
What it does well
- Automatic runtime detection scans pod metadata and processes to apply matching language instrumentation without manual intervention.
- Compiled language support uses eBPF to trace Go applications without source modifications or wrapper scripts.
- Declarative pipeline management uses Kubernetes custom resources like
SourceandDestinationfor GitOps workflows. - Collector scaling dynamically adjusts gateway and node collector replicas based on cluster telemetry volume.
- Granular telemetry controls mask sensitive PII, drop unneeded health-check spans, and normalize database queries before export.
