For small manufacturers
Ask your own data a question, in plain English.
SummaBrief is a desktop app for controllers and operations managers. It pulls the data out of the systems you already run, keeps a copy on your computer, and answers questions in a chat. Every number comes from code run over your data, never from the AI.
The app
One question at a time, answered from your records.
Point SummaBrief at your ERP, your warehouse or a folder of exports. It extracts what matters into a local copy. From then on you type a question and the answer comes back in the conversation: a figure, a table or a chart, with the program that produced it a click away.
How much did we scrap in June, and how does that compare to May?
June scrap was 1,284 lb, 2.1% of output. May was 1,912 lb, 3.4%. Most of the drop is on Line 2, where rejects fell from 41 to 12 build records.
Questions kept as templates
A question worth asking every week is saved as a template. Running it starts a new chat and computes the answer again, against the data as it is now.
Charts and print
Ask for a chart and it is drawn from the computed values. A conversation prints to PDF when someone needs it on paper.
Your data, on your computer
Answers are computed on your own machine, from an encrypted copy of the data that opens only with your login.
How it works
The AI writes the program. The program writes the number.
A general chat assistant asked about scrap gives a confident figure from nowhere. SummaBrief cannot: a number reaches the screen only as the output of code that ran over your data.
You
You ask
“How many pounds did Line 1 run yesterday?”
AI
The AI writes a program
A short piece of code against your extracted data. It is not allowed to answer with a figure of its own; a reply that states one is refused and asked for again as a program.
Code
The program runs
On your computer, against your data. What it prints is the answer. The AI’s own prose around the code is thrown away.
You
You see the result, and the program
The number, the table or the chart, and one click to the code that made it.
We do not say “verified” or “no hallucinations”. We say: computed from your data, traceable to the program that made it, and marked until a person has reviewed it.
Computed from source
Every figure is the output of code run over an extract of your own records. The AI never types a number into an answer.
Traceable
The program behind an answer is one click away, so a colleague who knows the data can see which tables and which filters produced it.
Marked until reviewed
Code can compute the wrong thing correctly. An answer nobody has checked says so, on screen and in print, until a person has looked at it.
Beyond the app
We also do the technical work around it.
Most manufacturers have more data than people to put it to use. We take on that work: the plumbing that gets the data somewhere it can be asked about, and the projects that come up once it is.
Data pipelines
Connecting to a warehouse, an ERP or a folder of exports, and building the extraction that keeps a clean, current copy of what matters. This is most of the work at any customer, and it is done before the first question is asked.
Models on production data
Machine-learning models trained on your own run history. For one plant we built a recommender that proposes machine settings for a target output, learned from years of controller records.
Setup and training
An audit of how your reporting works today, if you want one. Then the setup, training for the people who will use it, and the ongoing subscription.
We are looking for companies to build with.
We work closely with a small number of manufacturers during this early period. You get our direct attention, and software shaped around how your business runs.
If that sounds like your company, get in touch.