ARUNDO EDGE · DATA FROM ASSETS BEYOND THE NETWORK'S REACH
Every signal off the asset.
None lost on the way.
Arundo Edge is the industrial edge gateway that connects to the control system where the asset is — OPC UA, Modbus, MQTT — runs computes and machine-learning models on the box, keeps a local historian, buffers every tag through every outage, and streams compressed, in the order you choose, to Azure, AWS, Google Cloud, any MQTT broker, or Arundo Foundation. Managed as a fleet from one console. In the field since 2018.
- 2018In the field since 2018.
- BufferedConnection down for a day? The queue holds, batches, and resumes — newest first or oldest first, per destination.
- gzipTags are bundled into messages up to the size the destination accepts — 250 kB on Azure IoT Hub — and gzip-compressed.
- 0 inboundno inbound port, no VPN into the plant network
The asset is where the network isn't.
The link is the weak part.
A vessel on a satellite link, a wind farm on a cellular modem, a compressor station with no connection at all. The connection drops for minutes or days; the equipment does not stop producing data.
Protocols don't speak cloud.
The data lives in OPC UA nodes, in Modbus registers in a byte order the vendor chose, in MQTT topics with their own JSON. Someone has to read it right, timestamp it right and get it off the box — without opening an inbound port into the plant network.
A gap in the record is a gap in the model.
A model that watches a pump cannot be fed from memory. If the outage hours are missing, so is the fault that happened in them.
Connect. Compute. Buffer. Stream.
Arundo Edge is one service with four jobs, configured from its own screen on the box or from Edge Manager. Each job is a step the data takes on its way out.
01 · Connect.
Browse the OPC UA server and pick the nodes; map Modbus coils and registers over TCP, UDP or serial, in any of twenty-four value types; subscribe to MQTT topics with wildcards and map the JSON fields. Anonymous, user or certificate authentication; the OPC UA security modes and policies; server or local timestamps.
02 · Compute.
Arithmetic and conversions as expressions on the box — triggered on every value, on change, on a timer or a cron schedule, with the history window each input needs. Models run as Docker containers pulled from your registry, health-checked, and invoked with the same tags.
03 · Buffer.
Every tag bound for the cloud is written to a disk-backed queue before it is sent. Connection down for a day? The queue holds, batches, and resumes — newest first or oldest first, per destination. The destination confirms every batch; an unconfirmed batch goes back in the queue.
04 · Stream.
Bundled, gzip-compressed messages to Azure IoT Hub, AWS IoT, Google Cloud or any MQTT broker over TLS — several at once, each tag routed where you say — or written to an OPC UA server. Throttled to what the link can carry, and probed before real data resumes after a drop.
Level 1 on The Six Levels of Autonomous Plant Operations (Arundo 2026) is point analytics: a model on one asset, where the asset is. A model on the box is exactly that, and for an asset with no link it is as far as autonomy reaches on its own. Level 2 is the integrated plant — which is why Edge streams into Arundo Foundation, where the same signals join the Domain — a complete digital twin of your operation.
THROUGH AN OUTAGE
Built for the link you actually have.
Streaming is where edge software is usually thin: a retry loop and a hope. Edge treats the link as unreliable by design, and every setting below is per destination.
An offline buffer, on disk.
The queue lives in the PostgreSQL and TimescaleDB database bundled with Edge, so it survives a reboot as well as an outage. No database configured? An in-memory queue takes over. The box counts the tags waiting, so a backlog is visible before it is a problem.
Newest first, or oldest first.
Each destination chooses its publish order. Last in, first out puts the live values on the dashboard the moment the link is back and fills the history in behind them; first in, first out keeps the record in sequence for a historian or an OPC UA server.
Compressed, bundled, sized to the destination.
Tags are bundled into messages up to the size the destination accepts — 250 kB on Azure IoT Hub — and gzip-compressed. The box records both sizes per batch, so you can see what the compression bought.
Throttled, and probed.
A cap on messages per second for metered links. After a drop, small guard messages measure the round trip, and real data resumes only when the link answers within three times its usual round trip, three times in a row.
Every batch confirmed.
A batch leaves the queue when the destination acknowledges it; otherwise it is released and sent again. Batches are cut to the destination's message limit, with timestamp and quality on every value.
Several destinations, one box.
Azure IoT Hub, AWS IoT, Google Cloud, MQTT brokers over TLS with client certificates, and OPC UA servers, enabled side by side. Each tag carries its own list of destinations, so a KPI can go to the cloud while its raw inputs stay on the box.
Write-back, when you want it.
The OPC UA output creates its variables under a root path on your server and writes computed values to them — a KPI or a model's output becomes a node the control system can read. Oldest first and uncompressed, by default.
Compute where the data is.
For an asset with no link, the box is the only place analytics can run. For every other asset, it is the cheapest place to turn a thousand raw signals into the ten that matter.
Expressions.
Math-style expressions over any inputs — scaling, unit conversion, a ratio, a running KPI. Each input can carry a history window; each compute runs on every receive, on change, when all inputs have changed, on an interval or on a cron schedule, in priority order and in groups.
Models in containers.
A model is a Docker image in your registry — Docker Hub, a private registry, AWS ECR. Edge pulls it, starts it with an HTTP or TCP health check, assigns the port, invokes it with the tags it declares and maps the response to output tags. Edge Manager watches each run: duration, errors, container CPU and memory, logs.
A local historian.
Tags you mark for warm storage land in the bundled TimescaleDB: raw series, events, and 30-second, minute, hour and day aggregates — min, max, mean, sum — kept 180 days by default, with retention and chunking you set. Models read it through the local API; exports go to CSV or straight to Azure Blob Storage.
The whole fleet from one console.
Arundo Edge Manager is the fleet console. Each Edge polls it over an outbound connection every few seconds — no inbound port, no VPN into the plant network — picks up its tasks and reports back.
Status at a glance.
Online or offline, CPU, memory, disk, unsent tags, network connections and processes per device — current and as history — with labels to group the fleet and notes per box.
Every tag, end to end.
For each tag: the last value received at the box, the last value sent to the cloud, and when. For each input and output: connected or not. Pick a tag and watch it live.
Configuration, as a task.
Pull a box's configuration into the console, edit inputs, outputs, tags and computes there, deploy it back. Compute templates roll one group of computes out to many boxes. Every task is queued, acknowledged and logged, and succeeds or fails in view.
Models and versions, remotely.
Deploy a container, invoke a model or a compute group on demand, change or revoke a licence, export a period of the historian to Blob Storage, or push a new Edge version to the box — all as tasks.
Alerts before the backlog.
Thresholds on CPU, disk and memory, on unsent tags and on an unsent-tag count that keeps rising, and on time offline. Notifications are acknowledged in the console.
Audit and tenancy.
Single sign-on; one tenant per customer with its own devices, registries and licences; an audit log of who did what. Customers that allow-list addresses get a fixed one to point at.
In the field since 2018.
Arundo Edge runs on crane vessels, bulk carriers and LNG carriers at sea, on wind and solar farms, on hydraulic-fracturing fleets and at vehicle-charging sites — on Windows and Ubuntu, on gateway-class hardware and on the customer's own. The largest installations stream more than 4,000 tags each and run over a hundred KPIs and several machine-learning models on the box; the newest read vibration sensors over Modbus at six samples a second. The same software, version 6, is still the one installed today.
2018
First installation at sea.
4,000+
Tags streamed from one box.
6 Hz
Vibration over Modbus, buffered and streamed.
The details, in writing.
Is Arundo Edge for you?
Operators with assets a network does not reliably reach — ships, remote sites, fleets of mobile equipment — and OT teams that need plant data in the cloud without opening the plant network. It runs on the hardware you have, Windows, Linux or macOS, installed by a wizard or from the command line; a gateway-class box with 4 GB of memory streams thousands of tags. The first PLC is connected in an afternoon; the fleet follows one box at a time. Licensed per box, managed from Edge Manager, and the front door of Arundo Foundation: the same tags arrive in its industrial timeseries datastore and join the Domain.