HotLoop Flow
Wire the plant. Review it like code.
HotLoop Flow is a visual flow engine for the plant floor, written in Go. You wire sensors, brokers, PLCs and databases together on a canvas, and it runs those wires in one static binary small enough for a Pi, with a scheduler that doesn’t fall over when a sensor starts talking faster than the thing reading it.
What that means on a bad night:
- Every queue has a ceiling, and every message that doesn’t fit is counted. A fast sensor in front of a slow historian backs up, blocks or drops by the policy you picked, and it shows up on a graph before it shows up as an OOM kill with nothing in the log.
- A historian going down for an hour doesn’t cost you an hour. Mark a wire
durable and when the node behind it stops taking messages they go to disk in
order, and they’re still there after a
kill -9. While it’s healthy, nothing is written. Store and forward was proven against a real InfluxDB that was stopped for ten minutes. - Every node runs on its own. One goroutine per node, so a Function node chewing on a big payload costs you a core, not the editor and every other flow on the box.
- The flow file is something you can review. It loads and saves byte for
byte, an edit to one property is a one-line diff, and every save is atomic with
three backups behind it. Every deploy is a record:
who, when, why and the exact bytes, and any two of them
diff node by node, in the API, on the
command line and inside
git diff. Any of them rolls back, credentials included. - It won’t start unlocked. No authentication is a startup error, not a
default. The
execnode ships switched off, the file nodes are fenced into the data directory, and credentials are AES-256-GCM under an Argon2id key.
It runs your existing flows unchanged, too. That’s further down, because it’s a feature, not the point.
Current release is 2.16.6. The image is ghcr.io/hotloop-io/hotloop-flow:2.16.6
for amd64 (arm64 follows in 2.16.7), the Helm repo is https://hotloop.io/hotloop-flow/, and
Deploying Flow has Kubernetes, Podman and Quadlet. It has 117 node types for messages and 39 for FlowLab diagrams. Home and lab use is free, so is use by schools, nonprofits and government, and
use through EmberNet, a partner platform, business use included, is unrestricted, under the HotLoop Source License Agreement.
The numbers
Section titled “The numbers”Absolute numbers only, so you can check them on your own hardware. Three runs each, and the range is what’s printed.
| Measure | Result |
|---|---|
| Binary, linux/amd64, static | 41.7 MB |
| Image, amd64 | 45.64 MB |
| Start to answering its first request | 0.05 s |
| RSS, idle | 16.4 to 16.7 MiB |
| RSS, peak under the HTTP load below | 26.4 to 28.5 MiB |
| HTTP, five-node flow, 8 connections | 13.0k to 15.3k req/s |
| HTTP latency p50 / p99 | 0.39 to 0.41 ms / 1.78 to 4.03 ms |
| Engine, five-node chain, no I/O | 2.20 to 3.11 µs a message, 3.9 KB allocated |
Know what these aren’t. The HTTP number runs through the whole HTTP stack on purpose, because that’s what a client of your flow sees. And under saturation every runtime queues and queueing owns the tail, which is why p99 wanders more than p50 between runs.
On a Raspberry Pi 5
Section titled “On a Raspberry Pi 5”The same flow and the same commands, 2026-10-22, on a Raspberry Pi 5 Model B
(8 GB, four Cortex-A76 cores) with
its clock capped at 2.0 GHz, running the published 2.2.0 arm64 image as a pod
with nothing else on the node but cluster DaemonSets. Three runs, range printed.
| Measure | Result |
|---|---|
| Start to answering its first request | 0.10 to 0.46 s |
| RSS, idle | 88.0 to 88.3 MiB |
| RSS, peak under the HTTP load | 40.3 to 40.5 MiB |
| HTTP, five-node flow, 8 connections, 30 s | 13.2k to 13.4k req/s |
| HTTP latency p50 / p99 | 0.48 ms / 2.72 to 2.78 ms |
| Engine, five-node chain, no I/O | 7.74 to 8.15 µs a message, 4.0 KB allocated |
| HTTP, the same command held for 600 s | 11.4k req/s, p50 0.58 ms, p99 3.00 ms |
The last row is the honest one for a box that stays loaded. The board hit its soft temperature limit under a minute in, throttled from about two minutes in, and sat at 83 to 86 C with the clock at 1.5 to 1.7 GHz for the rest. The benchmarks page has the Pi run in full.
The benchmarks page has the method and the caveats.
Running it at a plant? Using HotLoop in a business covers licensing and how to reach us.