FoundryNet / Solutions / Legacy protocol integration

LEGACY & BROWNFIELD · PROTOCOL COVERAGE

RS-232 to OPC UA. Your Equipment Already Speaks. Forge Translates.

14 protocols·Serial to OPC UA·Read-only·No new hardware

You do not need to replace a 1995 Modbus PLC to put an AI agent on it. The same kernel that reads a 2025 OPC UA server reads the serial link on the machine next to it.

Get Forge on AWS Marketplace →See pricing →

Read-only. Forge reads through standard industrial protocols. Nothing is written back. Your equipment is never at risk.


The problem

The rip-and-replace quote is the real blocker

Around 70% of installed manufacturing equipment is 20+ years old. It is depreciated, it is running, and it is making money. The standard vendor answer — new controllers, new gateways, a capital project and an 18-month schedule — is why most brownfield AI programmes die in the budget meeting rather than the technical review.

But the equipment is not silent. A 1995 machine with a Modbus RTU serial link is already emitting everything an agent needs. Nobody translated it, because translating it was worth less than the integration cost. That calculation changes when the translation is a single API call.

70%Equipment 20+ years old
14Protocols supported
$0New hardware required
30sTime to first normalized read

Protocol coverage

All 14 protocols, one data layer

MTConnect
CNC / machine tool standard
OPC UA
modern PLC + controller standard
PROFINET / S7
Siemens automation networks
EtherNet/IP
Rockwell / Allen-Bradley + robots
Modbus TCP
the universal PLC fallback
Modbus RTU
RS-232 / RS-485 serial, 1979-present
ROS 2 (DDS)
robots and AMRs
J1939 (CAN)
heavy vehicles and off-highway
MQTT Sparkplug B
brokered IIoT telemetry
BACnet/IP
building and facility systems
SNMP
network gear and power infrastructure
MAVLink
drones and unmanned platforms
HTTP / REST
OEM cloud APIs
CSV
historian exports and batch files

Serial is not a second-class citizen here. Modbus RTU over RS-232 / RS-485 lands in exactly the same canonical schema as OPC UA, so a 1995 PLC and a 2025 controller produce comparable fields in the same response. The agent cannot tell which machine was older, and it should not have to.


Retrofit path

What a brownfield connection actually looks like

1 — Read what is already there

Point Forge at the existing serial link, PLC register map, MTConnect agent or historian CSV export. No controller firmware changes, no new PLC logic, no downtime window.

2 — Normalize on the way through

Raw registers and coil addresses become named canonical fields. 40001 stops being a number and starts being spindle_speed_rpm.

3 — Validate before the agent sees it

Six read-time validators catch stuck sensors, dropouts and impossible rates of change. Legacy equipment produces more of all three, which is exactly why this layer exists.

4 — Hand it to any agent

REST or MCP. The same 32-tool surface serves the 1995 machine and the 2025 one.


Old and new, same response

A serial register and an OPC UA node, normalized identically

Raw source inCanonical field outNotes
modbus rtu   40001: 3400spindle_speed_rpm: 3400holding register, RS-485 serial
modbus rtu   40012: 785spindle_temperature_c: 78.5scale factor resolved at map time
opc ua   ns=2;s=Spindle.Speedspindle_speed_rpm: 3400modern controller node
j1939   SPN 190: 1850engine_speed_rpm: 1850CAN bus, heavy vehicle

The scale factor case is the one that quietly ruins retrofit projects. A register holding 785 for 78.5°C is correct in the PLC and catastrophic in an agent that reads it as 785°C. Unit and scale resolution is part of the mapping, not an exercise left to the reader.

Get Forge on AWS Marketplace →See pricing →

Related

Keep reading

Mixed fleet normalization

Old equipment is rarely all one brand. Normalizing across vendors and vintages.

The missing data layer

Why integration cost, not model quality, is what kills industrial AI.

Predictive maintenance

Prediction on legacy assets, without the six-month integration project.


From the newsletter

The Machine Agent

Longer-form writing on industrial AI infrastructure by Foundry Labs.

The Tower of Babel

The most expensive unsolved problem in industrial AI

Read on Substack →