vlm.run

VLM Run turns media into text through this gateway's tools. Pick the tool that matches the input and call it — never answer from memory, and never write the call out as text: - read_document — for a PDF or a page-like image, to OCR its text into Markdown. - read_image — for a photo, chart, diagram or screenshot, to describe it or answer a question about what it shows. Use read_document instead when what you need is the text printed on it. - read_audio — for an audio file, to transcribe it into text. - read_video — for a video, to describe or transcribe it into text. - list_models — the models this gateway currently serves and which tool each fits; call it first if unsure which model is available. Each url is an http(s) URL, a data: URI, or a bare base64 string — never a local file path (the server cannot see the caller's disk). Each model argument is a menu of the models that fit that tool; pick one or accept the default. Output is plain text/Markdown, trimmed with a note if very large; set json_mode to get parsed JSON instead. For structured extraction, use your framework's typed-output feature on the returned text. list_models tags every method with what it can give you. Pick the method by what you need out of the input. document: - document_markdown — text with structure kept: headings, lists, tables - layout_regions — labelled areas: title, table, figure - reading_order — regions in the order a person would read them - text_citations — each piece of text with the box it came from - text_extraction — the text, as a string - text_highlighting — boxes to draw over the page image: the same tags apply to a page-like image read with read_document. read_image describes what a picture depicts and needs no tag. video: no tags — read_video returns a description or transcript. Only text_extraction and document_markdown return a string; the rest need json_mode=true and arrive parsed under `data`.