Open source

We build in the open.

The infrastructure we run depends on open source, so we contribute back: to the OpenTelemetry Collector, to bare-metal Kubernetes networking, and to telemetry collection with eBPF. Everything listed here is public and Apache 2.0 licensed.

View on GitHub ↗ See how it fits into Day 2

01 — Projects

Three projects, all upstream.

  • OpenTelemetry · Collector contrib

    Isolation Forest processor

    Inline, unsupervised anomaly detection for any OpenTelemetry Collector pipeline: traces, metrics or logs. It learns normal behaviour from recent telemetry and tags, scores or drops anomalies in flight, with no external ML service. It ships in the official contrib distribution (alpha), and our team co-owns it.

    Processor on GitHub

  • Kubernetes · Bare metal

    Cosmolet

    A Kubernetes controller that advertises service IPs over BGP for bare-metal clusters. It runs as a DaemonSet with direct FRR integration, advertises only healthy services, and uses leader election for high availability. Beta, with testing at scale in progress.

    Cosmolet on GitHub

  • eBPF · OpenTelemetry

    Telegen

    One eBPF-native agent for every signal: metrics, traces, logs, profiles and security events, discovered automatically across more than 100 protocols and platforms, with no SDK, sidecar or code change. It exports OpenTelemetry (OTLP) and Prometheus remote write.

    Telegen on GitHub

02 — OpenTelemetry

Inside the OpenTelemetry project.

Our team includes a member of the OpenTelemetry organisation on GitHub (@aarvee11), who is a code owner of the Isolation Forest processor in the Collector’s contrib repository.

We have also proposed a connector that evaluates alerting rules in real time on streaming OpenTelemetry signals, so alerts fire inside the pipeline instead of after the data lands in a database.

Read the connector proposal

03 — Why open

Infrastructure software should be inspectable.

The teams that run sovereign infrastructure need to see what runs on it. Open code lets them audit it, extend it and avoid lock-in to any one vendor, ours included.

Each one solves a problem we meet running AI infrastructure: collecting every signal without touching application code, spotting anomalies while telemetry is still streaming, and keeping bare-metal services reachable without a cloud load balancer.

Talk to us

Tell us what you need to run.

Every agreement is shaped to the workload: seats, sessions, capacity, term and location.

Talk to us hello@mirastacklabs.ai