mutexer Blog
Engineering and product writing on AI-assisted industrial software, private cloud inference, and the infrastructure that keeps mission-critical systems running. Written by the team building it.
Four open source EtherCAT master stacks, one honest comparison. The most mature one is IgH EtherLab and we could not ship it, for reasons that have nothing to do with the code. Here is the question that decides your stack before you write a line, and the class of device that quietly comes up empty.
Almost every benchmark for a local model measures tokens per second. For an agent working through a long task that is close to the wrong number. What decides whether it is usable is what the eleventh tool call costs, and that comes down to whether the prompt cache survives the tool loop. Here is what we measured, and the rule we ended up writing down.
██████╗ █████╗ ███████╗██╗███████╗
██╔═══██╗██╔══██╗██╔════╝██║██╔════╝
██║ ██║███████║███████╗██║███████╗
██║ ██║██╔══██║╚════██║██║╚════██║
╚██████╔╝██║ ██║███████║██║███████║
╚═════╝ ╚═╝ ╚═╝╚══════╝╚═╝╚══════╝We are releasing Oasis CLI Harness. Other AI coding agents run on your laptop and can only see files, which is half the job in a plant. This one installs on the industrial PC itself: Modbus, OPC UA and EtherCAT are tool calls, it can measure whether the machine is fit for real-time control, and every write is gated, sandboxed and logged.
Magnum 1 is mutexer's private cloud AI inference engine for industrial code generation. It runs on open-weight foundation models inside your own environment, so your code and prompts stay in your private cloud. Here is what it is, why we built it, and how it measures up.