How it holds up

The Edgepacer agent

Footprint, disk-backed buffering, guaranteed delivery.

Edgepacer is the agent at the edge of your estate. It's a single static binary written in Rust, deployed on a host or as a DaemonSet or sidecar, and it does the work you'd otherwise own yourself: find the sources, ship the data, and never lose any of it on the way.

Light enough to forget

In our own running fleet Edgepacer idles at under 0.1% CPU and roughly 10 MB of memory. There's one process to account for, not a collector pipeline with its own resource budget and failure modes. Parsing, structuring and storage all happen downstream, so the agent stays small by design.

It never drops a line

Delivery is guaranteed, backed by on-edge, disk-backed buffering. When the network blips or the backend is briefly unreachable, Edgepacer buffers to local disk and replays when the path comes back — data isn't silently lost in the gap. For a system of record that's the difference between a complete audit trail and a mostly-complete one.

It discovers, you don't configure

The agent finds containers, journald and systemd units, and log files on its own, and keeps that view current as services change. Kernel-level discovery through eBPF for network and source visibility is being built out, widening what Edgepacer can see without any per-host configuration.

Your compute stays yours

That idle footprint isn't a vanity metric — it's the point. Plenty of collection agents sit in the hundreds of megabytes of memory at rest, and some want gigabytes. Multiply that across every node in every cluster and it's real money, spent on the thing watching your applications instead of on the applications themselves. Edgepacer is built the other way around: it stays small so the compute you pay for stays with your workloads.

In our own internal testing, a single agent sustained roughly 500,000 log lines per second — about 100 MB/s on the wire — on under one CPU core, with a small memory footprint in normal operation and rising only under extreme, sustained load. The headroom that buys stays with the work that earns you money.

Internal numbers

These are measurements from our own profiling harness, not production-at-scale figures. Once we've measured Edgepacer in production at scale, we'll publish those numbers properly — including exactly how we took them.

edgepacer · status connected
agent cpu <0.1%
agent memory ~10 MB
active streams 4
delivery guaranteed