I wanted to try mapping telemetry ingestion to the bare metal. What started as an exploratory project turned into an explorartory product (maybe). I built aacyn, a single-binary engine that drops onto a $600 consumer Mini PC and confortably ingests and queries 5 million events per second without a cloud cluster.
Some info about how it works:
- the memory model: It writes directly to a file-backed mmap ring buffer using a strict Structure-of-Arrays (SoA) layout. You get a fixed disk footprint, safe wrap-arounds, and zero GC pauses.
- vectorized queries: Full-table scans are natively executed in C using AVX-512 SIMD instructions (~286μs to scan 5M events).
- drop-in ingestion: It speaks native OTLP (port 4318) via a lightweight FFI bridge. No custom SDKs—just point your existing OpenTelemetry collector at it.
- declarative filtering: A TOML file compiles into binary C-structs to drop noise or rollup metrics directly during the ingestion loop with virtually zero overhead.
ok, so as for the business the model: The core engine, the unthrotttled 5M event/sec throughput, OTLP ingestion, and a local WebGPU dashboard are 100% free for individual developers.
Then if you want to use it for a team, hooking it into your central Grafana instance, enabling Linux eBPF auto-discovery, or streaming cold-storage backups to S3/R2—there's a $4/month license key (for now).
This (I hope) funds my time maintaining the enterprise features without artificial capacity limits.
You can spin it up in a few seconds: curl -sSL https://aacyn.com/install.sh | bash
We just had a sprint from version 0.5 to 0.7 overnight, so there might be some small bugs. Feel free to email feedback@aacyn.com
Thanks