HiveMQ
HiveMQ Edge: The Open-Source Gateway from PLC to MQTT
HiveMQ Edge is an Apache 2.0 gateway that turns PLC data into MQTT at the plant edge. What is free, what needs a licence key, and how it fails.
HiveMQ Edge is a gateway that reads data from PLCs and other plant devices, publishes it as MQTT, and can bridge it to a broker elsewhere. HiveMQ describes it as an open-source MQTT gateway under the Apache 2.0 licence, with a set of commercial features that switch on with a licence key. As of 20 November 2026 the current release is 2026.16, published 10 November 2026, and it runs on JDK 25.
This page covers what Edge does in a plant, what is free and what is not, how to decide whether it belongs in a design, and how it fails. It sits under the HiveMQ technology hub, and the company behind it is profiled at HiveMQ the company. For where Edge fits in a namespace design, see HiveMQ and the Unified Namespace.
What HiveMQ Edge does in a plant
It moves values from a controller to a topic. Edge has an adapter for each device protocol. You define tags on an adapter, and Edge turns each tag into MQTT messages on a topic you choose. A local MQTT broker is built into Edge, so other clients on the line can subscribe to those topics without going to the cloud. A bridge then forwards selected topics to a broker at the plant, the site or the enterprise.
Why that matters on a plant floor. The usual alternative is a point-to-point integration between every consumer and every PLC. Each consumer polls the controller on its own schedule, and the load on the PLC grows with the number of consumers. With a gateway the PLC is read once, and the consumers read the broker. That is the case for any edge gateway, and Edge is one implementation of it. For the wider pattern see edge solutions in industrial automation.
HiveMQ Edge open-source and commercial features
One package, two tiers. HiveMQ’s documentation states that Edge ships as a single package containing both open-source and commercial features, that open-source features are free, and that commercial features require a licence key. HiveMQ released Edge as open source in July 2023 and announced general availability with the commercial features on 19 March 2024. Those dates come from HiveMQ’s own news and documentation.
| Capability | Tier as documented (20 November 2026) | What to check |
|---|---|---|
| Protocol adapters: OPC UA, Modbus, Siemens S7, Beckhoff ADS, EtherNet/IP, BACnet, HTTP(S), File, Simulation | Open source | Most are read-only toward MQTT |
| MTConnect and Databases adapters | Documented, but not in the open-source list | The licence tier is not stated; ask HiveMQ |
| MQTT 3.1, 3.1.1 and 5 over TCP, TLS and WebSocket | Open source | Local broker listener settings |
| Bidirectional MQTT bridging | Open source | Buffering during an outage is separate |
| Topic prefixing for the ISA-95 hierarchy | Open source | You still design the topic tree |
| Java extension SDK | Open source | Custom adapters and interceptors |
| Offline buffering (store and forward) | Commercial | Sizing and eviction are undocumented |
| Data Hub on Edge (schemas, validation, transformation, behaviour enforcement, REST API) | Commercial | No trial is offered on Edge, per HiveMQ’s Data Hub documentation |
| Southbound MQTT to OT (writes to devices) | Commercial | Overview names the OPC UA adapter only; EtherNet/IP write options are also documented |
| Enterprise support | Commercial | Terms are not on the docs page |
The line is easy to cross by accident. A design that needs a write from MQTT to a PLC, or that must not lose data over a long link outage, needs the licence. An older HiveMQ community answer says only offline buffering, Data Hub and support need one. The current documentation adds southbound writes, so we treat the documentation as authoritative. For what the licence costs, see HiveMQ editions and pricing.
How HiveMQ Edge is selected
The decision is usually one of three shapes.
- A gateway at the machine or cell. You need PLC, OPC UA or Modbus data as MQTT with little code. Edge is a fit when the protocols you need are in its adapter list. Check the list first: HiveMQ’s documentation we read names no adapter for PROFINET, and no native tag-name adapter for Rockwell Automation controllers other than EtherNet/IP addressing by CIP class, instance and attribute. We could not confirm these as absent across the whole page, so verify against the live page.
- A broker at the plant. If many clients connect and the broker must survive a node loss, you want a clustered broker, not a gateway. Edge can still sit below it. See HiveMQ clustering and high availability.
- A third-party gateway feeding any broker. If you already run a gateway that speaks MQTT, adding Edge duplicates it. Our view: do not add a second hop for its own sake.
Questions we would ask before choosing Edge:
- Which protocols and which controller families, and are those adapters read-only, write-capable or commercial?
- How many tags per device and how fast, given that every polled tag is load on the PLC?
- What must happen to data when the link north is down, and for how long?
- Who owns the topic tree, and is it written down before the first adapter is configured?
- Which Java and operating system will you run, and who patches it?
How HiveMQ Edge fails
HiveMQ’s documentation describes behaviour; it does not catalogue field failures. The table below mixes documented settings with our reading of them. Items marked as our view are inferences.
| Failure mode | Why it happens | What to do |
|---|---|---|
| PLC load from polling | Polled adapters default to a 1000 ms interval and accept down to 1 ms, per HiveMQ’s adapter documentation | Start slow, count tags, read the controller’s scan and comms load |
| Tag sprawl (our view) | Tags are easy to add in a browser console, and no maximum is documented | Review tag lists like a change request; use the bulk tag operations deliberately |
| Wrong values from data types | Modbus returns 16-bit registers, so multi-register types need register flipping; EtherNet/IP needs the right data type and byte order | Test every tag against a known value at the device |
| Timestamps (our view) | OPC UA payloads carry source and server timestamps when metadata is present; we did not find what time a polled-adapter payload carries | Test with a deliberate change and compare the clocks |
| OPC UA reconnects | HiveMQ’s knowledge base has an article titled “HiveMQ Edge OPC-UA: Data Flow Stops After Reconnect”; we could not read it | Pin a release after reading the changelog; test pulling the network cable |
| Buffer exhaustion | Offline buffering is commercial; HiveMQ does not document sizing, and with in-memory persistence a restart loses queued messages | Use file-native persistence in production and test an outage of the length you must survive |
| Silent upgrades (our view) | Edge shipped seven releases between 16 June and 10 November 2026 | Pin versions, read each changelog, upgrade a test unit first |
Polling is the failure we would watch first. The point of a gateway is to protect the controller, and a bad polling plan defeats that. The reasoning is in PLC scan cycles, polling and SCADA data. The adapter details are on HiveMQ Edge protocol adapters.

Persistence is a decision, not a default. HiveMQ’s configuration documentation says Edge stores sessions, queued messages, retained messages and subscriptions in memory by default, that this data is not written to disk when Edge stops, and that it recommends file-native persistence for production. It also says only QoS 1 and QoS 2 payloads are persisted. A line that publishes at QoS 0 and then loses its link therefore loses data by design.
Lifecycle and licensing
Releases are frequent and numbered by year. HiveMQ published 2026.10 on 16 June 2026 and 2026.16 on 10 November 2026, with five releases between. Edge runs on JDK 25 from 2026.8; earlier, 2025.16 needed Java 21 and HiveMQ advised against any later version on 32-bit architectures. We found no long-term support line for Edge comparable to the broker’s, so for a plant, pinning and a test unit stand in for one. HiveMQ’s 2026.16 notes list, as a known limitation, at most 30 to 40 protocol adapters with southbound mappings in one Edge, which is the only capacity figure we found. For the broker’s lifecycle, see the HiveMQ Platform page.
What we do here
We design MQTT and Unified Namespace architectures and integrate brokers with PLCs, SCADA, historians and enterprise systems as part of IT/OT projects. We are independent of HiveMQ: we have no commercial relationship with the company and do not resell its products.
- Gateway placement: deciding whether a gateway belongs at each cell, line or only at the plant, and where it sits on the network, with OT network segmentation in mind.
- Tag and topic design: writing the topic tree and tag list before configuration, so the namespace is not an accident of adapter order.
- Polling budgets: setting read rates against what the controller can give without hurting the process.
- Failure testing: specifying and witnessing link-loss, restart and reconnect tests before go-live.
- Licence questions: reading what a design needs from the commercial tier before the quote.
IT/OT integration is the service this work sits under.
Limits and open questions
- Which adapters are commercial beyond southbound writes. The MTConnect and Databases adapters are not in the open-source list and their tier is not stated.
- Offline buffer sizing and eviction for the Edge-level buffer are not documented on the pages we read. The bridge page documents a queue limit and a discard policy for bridges; see bridging.
- Whether Edge can publish Sparkplug B. We found no Sparkplug adapter in the documentation we read. Treat it as unconfirmed.
- The Pulse agent. Edge 2025.16 added a Pulse Agent, and the 2026.16 notes mention a Data Intelligence agent. We could not confirm whether one was renamed into the other.
- Platform and Java. Confirm the supported operating system and runtime against HiveMQ’s system requirements page before you install.
- Parameter names and defaults here come from pages read on 20 November 2026 and change with releases.
Deeper
- HiveMQ Edge protocol adapters: each adapter, read-only against write-capable, polling and subscription defaults, and the S7 PUT/GET requirement
- HiveMQ Edge-to-cloud bridging: topic mapping, loop prevention, QoS and what happens when the link drops
Across to a sibling: HiveMQ and the Unified Namespace. Back to the HiveMQ hub or the company profile.
Frequently asked questions
Is HiveMQ Edge free?
Partly. HiveMQ's documentation states that open-source features are free to use under Apache 2.0 and that commercial features need a licence key. The free list includes the protocol adapters, MQTT bridging and ISA-95 topic prefixing. Data Hub on Edge, offline buffering, southbound MQTT to OT and enterprise support are commercial, as of November 2026.
What does HiveMQ Edge connect to?
HiveMQ's documentation lists adapters for OPC UA, Modbus TCP, Siemens S7, Beckhoff ADS, EtherNet/IP, BACnet/IP, MTConnect, HTTP(S), files, databases and a simulator. Most of the PLC adapters are documented as read-only toward MQTT. HiveMQ's overview calls southbound MQTT to OT a commercial feature, currently for the OPC UA adapter only; the adapter page also documents write options for EtherNet/IP logical addressing.
Does HiveMQ Edge need Java?
Yes. HiveMQ's documentation states that Edge runs on JDK 25 from release 2026.8, and that container image and Helm chart users need no action. People installing from the ZIP package should check the system requirements. Edge 2025.16 required Java 21 or higher, so older installs may need a runtime change before upgrading.
Does HiveMQ Edge keep data when the cloud link drops?
Only with the commercial offline buffering feature. HiveMQ's documentation says bridge buffering is configured per bridge, holds QoS 1 and QoS 2 messages, and survives a restart only in file-native mode. HiveMQ does not document how much data a buffer can hold for a given disk size, so size and test it yourself.
Should I use HiveMQ Edge or a HiveMQ broker at the plant?
Edge fits a line or cell where the job is reading PLCs and forwarding to MQTT, because it carries the adapters. A full HiveMQ broker fits when many clients need a clustered, highly available broker. Many designs use both, with Edge at the machines bridging to a plant or central broker.
References and further reading
- HiveMQ Edge documentation · HiveMQLicensing, the open and commercial feature lists, platforms and the JDK requirement.
- HiveMQ Edge protocol adapters · HiveMQAdapter settings, defaults and the southbound rule.
- HiveMQ Edge MQTT bridges · HiveMQ
- HiveMQ Edge configuration · HiveMQIn-memory against file-native persistence.
- HiveMQ Edge 2026.16 release notes · HiveMQThe current release at 20 November 2026.
- HiveMQ Edge adds data transformation and enterprise-grade reliability to bridge OT to IT · HiveMQThe 19 March 2024 general availability announcement.
- hivemq-edge repository · HiveMQ on GitHub