Root field `pdf_preprocess`

When POST /v1/chat/completions messages contain a parseable PDF (validated input_file in a user message, normalized internally to a file part with file_data / file_id / file_url), this service may preprocess the PDF before forwarding upstream.

Default behavior (important)

Omit root pdf_preprocess, or pass an object without engine: both mean native—no expansion; keep the file part for upstream (model must support native PDF input).
engine only fails validation on invalid values (400 / invalid body).
false / null: disables the preprocess config object (exact semantics vs “omit object” follow the README).

`engine` values

`engine`	Behavior
`native`	No expansion; keep normalized file part.
`ocr`	Replace the PDF part with per-page JPEG `image_url` (data URLs); no `[PDF]` meta line or full text layer; user `text` in the same `user` message is kept.
`markdown`	Meta `text` + Markdown-style `text` from the PDF text layer (no complex layout fidelity).

Optional parameters

max_pages: defaults and caps in pdf-preprocess/constants (README summary; typical default 10, cap 50 order of magnitude).
merge_pages: when true, affects ocr rendering only (merge pages vertically into one JPEG).

Runtime and ops

WASM: env GATEWAY_PDFIUM_WASM_PATH points to pdfium.wasm; if unset, try @embedpdf/pdfium package path (depends on cwd—set explicitly in production).
Preprocess failures may return 4xx / 504 with error.code such as pdf_password_required, pdf_expand_too_large, etc.—full list and size/timeout limits in the README and constants source.

SDK mapping

Put config in chatRequest.plugins.pdf; @routerbrain/sdk serializes it to JSON root pdf_preprocess—see SDK: chat.send.

Root field pdf_preprocess