Your BOM Is Not the Price of Admission

Every LCA practitioner knows the moment. A tool asks you to upload the file — the BOM with your supplier’s part numbers, the cost sheet procurement spent two quarters negotiating, the spreadsheet that, read sideways, is a map of who you buy from and what you pay. Your cursor sits on the button. You remember the contract clause legal flagged. You close the tab and go back to doing it by hand.

That hesitation is sound. Most LCA software treats your supply-chain data as the price of admission: hand it over, and the tool will tell you something about it. The trade is so routine that nobody names it anymore. But the BOM was never yours to give away. It is the supplier’s part numbers, procurement’s pricing, the competitive shape of how your product gets made.

The trade nobody names

Tools ask for the upload because of where they run, not because anyone wants to take it. The reasoning lives on their servers, and to reason over your file, the file has to be where the reasoning is. So up it goes.

For a material name typed into a search box, fine. For a 400-row BOM where column C is supplier and column D is unit cost, not fine. These are not the same kind of data — but the upload button can’t tell them apart.

The industry’s reflex is to answer with “secure.” Encrypted in transit. SOC 2. A data-processing addendum. All real, all beside the point. “Secure” describes how well your data is guarded once it has left your machine. It never touches the prior question: did it need to leave at all?

Where each thing happens

Cortex Cowork answers that question by drawing one clean line through the work, and putting each half on the side that should hold it.

The cloud does the reasoning: planning the task, deciding what to look up, searching across fourteen LCA databases — HiQLCD, Ecoinvent, EF, CarbonMinds, and others. That work belongs next to large models and a maintained index. It also needs to know nothing about your supplier.

Your computer does everything that touches your data. Cortex Cowork runs locally — macOS on Apple Silicon or Intel, Windows, Linux — with nothing to configure. It reads and writes the files in your project folder, searches them, runs local tools, and keeps the workspace and conversation history on disk. Your BOM, your cost sheet, your supply-chain map: read in place, never sent up.

Think in the cloud, act locally. The reasoning comes to your data. Your data does not go to the reasoning.

What crosses the line is the question — “find a GWP100 factor for 304 stainless, EU, cut-off” — not the row it came from. Cortex learns what to look for without ever seeing the file the row lives in.

The snail keeps its shell

A snail doesn’t mail its shell ahead and travel light. The shell goes where the snail goes; the two don’t separate, and that’s the whole point of having one. Your project data is your shell. Cortex is built so the data stays attached to you and the intelligence travels to meet it.

In practice, the things you’d worry about stay put. Files never leave the device. Project memory — the decision log, the proxy substitutions you signed off on in week one, the client conventions in the project wiki — lives locally. Cortex does not federate data across customers; nothing you do in one project surfaces in anyone else’s. The next practitioner who opens the workspace sees your reasoning because it sits on the same disk, not because it round-tripped through a shared server.

It still drives your tools

Local is not the same as isolated. The machine that holds your BOM also runs your modeling engine, and Cortex works there too. It connects to and operates openLCA, brightway, and 积木LCA: matching background datasets, building the product system, running the calculation. It doesn’t replace the engine — it drives it. You stay in the tool you already trust, and the engine never has to reach across a network you can’t see.

Where automation would quietly break an audit, Cortex stops and hands the decision back. Coverage too thin to defend. A proxy that’s close but not the right region. The same material returning factors that differ by more than 2× across databases. A restricted dataset you don’t yet have access to. You decide; the decision goes into the reasoning chain. None of that needs your file to be anywhere but in front of you.

Local, on purpose

The upload button asks you to trade supply-chain data for an answer. Hesitating was always the right call. The better arrangement is the one where the trade never comes up: the intelligence comes to your data, works beside it, and leaves the file where it started — on your machine, in your folder, under your control.

Slow is fine. Local is the promise.

Try it on a real BOM — download Cortex Cowork for macOS, Windows, or Linux, or write to us at lizj@hiqlcd.com.