Document parsing

Most "document AI" tools either flatten everything to plain text (losing structure) or bolt on a different pipeline per format (multiplying bugs). ContextBoard does neither: every adapter -- PDF, DOCX, PPTX, XLSX, CSV, TXT -- produces the same shared block-based document model, so every downstream feature (search, extraction, viewing, context packs) works identically regardless of what the file started as.

The document model

A parsed document is a sequence of typed blocks: headings, paragraphs, lists, tables, images, code, quotes, footnotes, and speaker notes, each carrying its position in the source (page number, section path). Nothing is lost to a plain-text flattening pass -- a table stays a table, a heading stays a heading.

Content-addressed by design

Identical bytes uploaded twice in a workspace are parsed once and stored once. Re-uploading the same file to correct a typo in its title doesn't trigger a duplicate parse or duplicate storage cost.

Built independently, against public specifications

The parsing pipeline's architecture -- content-sniffing format detection, per-format adapters, a shared document model, and pluggable renderers -- follows publicly documented patterns common across the document-parsing ecosystem. Every adapter is written independently against the relevant public format specification (OOXML, RFC 4180); no third-party parser source is vendored or copied.