First Clef, now Clairic.

Clef picks from options you give it. Clairic extracts values from a receipt photo, a support email, or a CSV of daily counts.

You get typed answers with confidence scores in one forward pass, without generating output tokens.

Use them ahead of an LLM and you can keep the raw input away from the model writing the reply. It gets the extracted values, not the original email.

Great work by @michellechen and the Workers AI team. In preview now, with open weights coming soon.