Skip to content

Find a part

Open a project, go to the Components tab, and type a part number.

Ranked candidates, each with its manufacturer, a datasheet link and a confidence signal. Nothing is chosen for you. Open a datasheet link before confirming if the match is not obvious — that is what it is there for.

If you typed something slightly wrong, you get a did you mean card rather than a silent correction. That is deliberate: a silent substitution that happens to be right is the same mechanism as one that is wrong, and next time it produces a part you did not ask for.

Parts already in your library are marked as such, so you do not re-add one you already own.

Confirming a candidate runs extraction against its datasheet and opens Part Detail with a real pin table — numbers, names, electrical types.

Every field records where it came from: which source, which model, and a confidence signal. That record follows the part for its whole life, so months later you can tell a value read out of a datasheet from one a model inferred.

Save to Library turns it into a real object on your disk — readable JSON you can open, diff and commit. It is now available in every project.

Reopen it any time by clicking it in the Library rail. That loads the saved record directly; it does not re-run extraction or cost you another API call.

A search that is too broad returns a short, unhelpful candidate list. Add the package or the manufacturer.

A dead datasheet link happens — manufacturers move files. Confirmation is not blocked by a failed datasheet fetch, so you can still save the part; you just will not get design guidance until a datasheet is reachable.

“Extraction did not return valid JSON” means the model’s response was malformed or truncated. Retrying often works. If it is reproducible for a specific part, that is worth an issue — include the part number.