Tsinghua roots, then ChatGLM
Zhipu AI grew out of Tsinghua University’s knowledge-engine work. ChatGLM made a bilingual assistant that felt native in 中文 long before that was common. The GLM-4 line pushed the house toward agents and structured output — models that would rather return fields than a paragraph. GLM-4.5 is the current deep seat. That instinct is why this rail exists: read the page, emit the fields, stop.
GLM-4 and GLM-4.5
GLM-4.5 is the careful extract. GLM-4 and its flash twin handle a short field list. On Wrooms the deliverable is the fields, not a chat. An English spreadsheet with lookups is Sheets. A 中文 government or ops form is GLM Form. Translate can move labels. It does not own the fields.
What GLM Form reads
A 中文 form, a scanned table, a page of labeled boxes. It reads what is there. It does not design a tool schema — that is house Tools. It does not rerank a search list — that is Rerank. It does not write a Kimi dossier.
How to ask for fields
Paste the form or name the columns. @glmform or pin GLM Form. Keep the room on extract.