Ask an industrial engineer how their plant would recover if a controller’s code got corrupted or encrypted by ransomware, and the honest answer is often grim: a backup on someone’s laptop, in a folder named “final_backup_2”. Copia Automation wants to fix this vulnerability by borrowing practices from software development: version control, collaboration, and structured code management.

The New York-based startup has just raised $26 million to build the “GitHub for the factory floor,” a platform designed for technicians working on PLCs, HMIs, and other control systems. Rather than adapting a general-purpose tool, Copia offers an environment that understands proprietary languages and machine logic, tracking every change with the same granularity that Git applies to web application commits.

The paradox of invisible code

At the heart of the problem lies the nature of software that runs production lines, robots, and presses. Often written by external system integrators or in-house technicians with vertical expertise, this code lives in isolation, without central repositories or disaster recovery procedures. In the worst cases, the only backup copies are scattered on USB sticks or unversioned network shares. If an attack halts operations, recovery can take days, with costs far exceeding any investment in protection tools.

Copia addresses this by providing a system that tracks changes, enables collaboration among multiple technicians, and allows immediate rollbacks. This is a cultural shift even before a technological one: moving from the logic of a hand-saved “final file” to the process automation typical of modern software pipelines.

On-premise by definition

From the perspective of those deploying critical infrastructure, the key aspect is that Copia is designed for on-premise environments. Many plants cannot or do not want to expose their control systems to the cloud, due to latency, security, or data sovereignty constraints. The platform installs locally on company servers and communicates directly with field devices. This architecture meets a growing need in the world of Large Language Models: when models run at the edge for applications like predictive maintenance or visual quality inspection, versioning of weights, training data, and inference scripts becomes as critical as traditional PLC code.

AI on the shop floor: the same rules as machine code

Industry is embedding artificial intelligence models directly on production lines, often on on-premise hardware such as edge servers or GPU-equipped workstations. These models – typically specialized LLMs or vision networks – share the same nature as controller software: they are artifacts that evolve over time, require audit trails, and must be recoverable after failures or attacks. A flawed update to an anomaly detection model can halt production just like a corrupted PLC. A structured versioning system for all software assets in the plant, whether ladder logic or quantized weights, reduces operational risk and simplifies compliance with regulations and audits.

For those evaluating on-premise deployment of AI solutions, the parallel is instructive: Copia’s focus on versioning OT code suggests that the same discipline should apply to local LLM inference, where model version control and pipelines are still often managed with homegrown procedures. AI-RADAR provides analytical frameworks at /llm-onpremise to explore these trade-offs without sliding into direct recommendations.

What’s at stake

The $26 million raised by Copia signals that the market recognizes the need for tools built for the complexity of industrial environments, not just data centers. In an era where ransomware regularly strikes critical infrastructure, the ability to quickly restore the code that moves machines is a resilience asset. And as factories become increasingly data-driven, the line between machine code and artificial intelligence blurs: both demand version control, secure backups, and verifiable deployments.