AI Copilot

Library check with reasons

Every BOM row is matched to a symbol and footprint — and a rejection names its rule.

/findings

The AI Planned BOM review on an NE555 blinker: each part checked against its symbol and footprint, two capacitors rejected because their 0805 footprint is not loaded, and the NE555's symbol and land side by side

A BOM row is only useful if it lands on the right symbol and the right land pattern. The library check is deterministic: it compares the part's package, pin count and pads with your libraries and, when it refuses a match, tells you which rule decided and why.

What it does

No model involved

Matching is code, not a guess, so it gives the same answer every time.

Pins versus pads

The symbol's pins are checked against the footprint's pads.

A reason, not "invalid"

A refusal names the rule — wrong package family, wrong pin count, missing pad.

Fix where it belongs

The result tells you whether to repair the part, the library or the BOM row.

Matching is code, not a model

The Copilot may propose a part, but whether it binds to a symbol and a land pattern is decided by deterministic checks that give the same answer every time and name the rule when they say no. That makes the result fast — milliseconds per row — and repeatable: run the same BOM twice and the same rows pass. The same checks run on a row you typed yourself and on one the Copilot planned.

Five checks, in order

A candidate is not accepted because its name looks right. It must be the same class of part (a resistor row does not take a connector), have as many electrical contacts as the row declares, have a pad with the same number for every symbol pin and none left over, fit the package the part really has, and not short pads of different signals or place them closer than 0.1 mm. When a check fails, the BOM card says which rule decided and why, so you know whether to repair the part, the library or the row.

Shorts that matter, and only those

Manufacturers join ground to an exposed pad on purpose, so two overlapping pads are an error only when the symbol says they are different signals. The audit was calibrated on all 15,447 footprints of KiCad's library: apart from footprints with no pads at all, none is flagged as an error. The same overlap between VIN and VOUT is a dead board, and is reported.

Where it looks, and in what order

First the pool: the project's own library, parts a person confirmed on earlier projects, and the libraries you have loaded — scanned for the files whose names fit the row rather than loaded wholesale. Only when nothing there passes does it ask WireFrame's published library of 10,000+ parts, then LCSC by the exact part number, then a generated draft. A real part beats a fabricated one, but a wrong real part is the worst outcome, because it looks right.

The details

ChecksClass · contact count · pins against pads · physical fit · copper shorts and clearance (0.1 mm)
CalibrationAll 15,447 footprints of KiCad's library
SourcesProject library, confirmed parts, loaded libraries, then the published library (10,000+ parts), LCSC by exact part number, a generated draft
Family namesKiCad family symbols such as SHT4x or LM78xx match the orderable part number
Model callsNone — /bom, /missing and /findings read the same data instantly

How to start

  1. Draft a design with the Copilot, or open an existing project.
  2. Open the BOM review and read each row's verdict.
  3. Repair a rejected row with /fix or the symbol and footprint editors.

Questions

Why was a part I know is right rejected?

The card names the rule. The usual reasons are a contact count that differs from the row's, a symbol pin with no pad of the same number, or a footprint that is not loaded. Fix the row, the symbol or the library accordingly, and the card is checked again on its own.

Is a part confirmed on one project trusted on the next?

It is reused, but judged again: the files join the pool and the new row must still pass every check, because it may ask for another package. Only parts a person confirmed are kept, never automatic acceptances.

Does LCSC take the closest match?

No. Only an exact part number is taken — AMS1117-3.3S is a different part from AMS1117-3.3 and is listed for you rather than taken — and the downloaded part is checked again before it is written.