Why this exists
You are putting a microcontroller on a board. You have its datasheet open — two hundred and thirty pages of it. Somewhere in there is the answer to what that part actually needs around it to work: how much decoupling and where, whether the reset line needs a pull-up, what load capacitance the crystal wants, what the brown-out behaviour does to your EEPROM if the supply sags.
That information exists. It is written down, precisely, by the people who made the part. And finding it means scrolling.
So you scroll. You find the power section, read three paragraphs, and go back to KiCad. Twenty minutes later you need the reset section, and you scroll again. By the end of the afternoon you have made four correct decisions and cannot remember which page any of them came from — which matters, because in six weeks when someone asks why there are two capacitors on that pin, “I read it somewhere” is not an answer you want to give.
Then the board is done and it needs a case. Your enclosure is a separate program, with a separate model, that has never heard of your board. So you read your own mounting-hole coordinates off the PCB editor and type them into FreeCAD by hand. If the board outline changes, you do it again.
None of this is hard. All of it is slow, and all of it is the kind of slow that produces mistakes — a transposed coordinate, a decoupling value remembered from a different part, a standoff in the wrong place discovered after the print.
What we thought was worth building
Section titled “What we thought was worth building”Three things, in the order they hurt.
Answers that carry their source. The value is not that a model can tell you about decoupling — it is that you can check. Every piece of design guidance in this app is extracted from your part’s real datasheet and stored with the page it came from. Click the citation, the PDF opens at that page. An item whose quote cannot be found on the page it claims is discarded rather than repaired. That contract is the whole feature; the plain-language summary on top is just what makes it readable.
One library, across projects. A part you looked up once should be a real object you own — with its pins, its package, its datasheet, and a record of where each field came from — not a chat message you scroll back to find. Save it once and it is there in every project after, along with the footprint, which is shared rather than duplicated, exactly as KiCad models it.
Geometry that comes from the actual board. Your enclosure should be built from your board’s real outline and real mounting holes, read from the file, not from numbers you retyped.
The thing we deliberately did not build
Section titled “The thing we deliberately did not build”Every few months something appears that claims to generate a schematic from a prompt. We are not that, on purpose.
The failure mode of a confident wrong answer in hardware is expensive and delayed: a hallucinated footprint that looks plausible costs a PCB spin, and you find out weeks later. So the app validates before anything reaches a board — pin count against the package, pad pitch for sanity, courtyard actually enclosing the pads — and refuses rather than guessing when a package is not recognised. Writing to a board you have open in KiCad happens only after you have seen exactly what will be written and clicked confirm.
The honest framing is that this is an advisor with hands. It reads, explains, cites, and — when you tell it to — writes one specific reviewed thing. It is not trying to design your board, and the moment it started trying, you would stop being able to trust the parts that work.