PDF

Rasterize PDF pages to images (PNG/JPEG/WebP) and author editable AcroForm fields at measured coordinates. Rendering uses pdf.js on a server-side canvas; field writing uses pdf-lib.

Why use this

• One coordinate space — both kinds speak pixels of the rendered image, top-left origin, at a render scale. Coordinates measured on a Pdf.Rasterizer image (by a vision model, an Image.Overlay preview, or a human) feed Pdf.FormFields unchanged; the conversion to PDF user space (points, bottom-left origin) happens inside the controller. The rasterizer reports the scale it rendered at, so the contract is wirable — pass result.scale into the Pdf.FormFields invocation instead of keeping two config values in sync.
• Bytes in, bytes out — both kinds take and produce buffered Uint8Array payloads, so they compose with S3.Get/S3.Put, Octet.Decoder, and HTTP bodies without touching the filesystem.
• Actionable failures — unparseable bytes, out-of-range pages, duplicate field names, and out-of-bounds boxes all raise ERR_INVALID_INPUT with the offending field and bounds spelled out.

Kinds

Kind	Purpose
Pdf.Rasterizer	Render one page of a PDF to image bytes (png/jpeg/webp) plus pixel dimensions and page count.
Pdf.FormFields	Add editable AcroForm fields (text, checkbox) to a PDF at rendered-image pixel coordinates.

Example

kind: Telo.Application
metadata: { name: form-stamper, version: 1.0.0 }
imports:
  Pdf: std/pdf@latest
---
kind: Pdf.Rasterizer
metadata: { name: Render }
scale: 2
---
kind: Pdf.FormFields
metadata: { name: AddFields }
---
# Render page 1, then place a text field using coordinates measured on the
# image — the render's scale is wired through, so the two can't drift.
- name: page
  inputs: { data: "${{ steps.fetch.result.bytes }}", page: 1 }
  invoke: !ref Render
- name: fielded
  inputs:
    data: "${{ steps.fetch.result.bytes }}"
    scale: "${{ steps.page.result.scale }}"
    fields:
      - { name: firstName, type: text, page: 1, x: 240, y: 64, width: 300, height: 40 }
  invoke: !ref AddFields