Skip to main content

​
Overview

When you upload a PDF or DOCX file, Raycaster Doc automatically parses it into structured Markdown in the background. This parsed content powers AI features like chat document reading, semantic search, and review analysis — without you having to do anything.

​
How Parsing Works

1

Upload triggers parsing

When a file is uploaded or replaced, a SHA-256 hash of the source file is computed.
2

Cache check

If parsed content already exists for this exact file hash, parsing is skipped entirely (cache hit).
3

Parse job queued

For new files, a background parse job is queued and processed asynchronously.
4

Parsed content stored

The output — per-page Markdown files and extracted media — is stored in a separate cache bucket.

​
Parsing by File Type

FormatParserOutput
PDFMistral OCRPer-page Markdown + extracted images
DOCXReductoPer-page Markdown + media
DOCReducto (legacy)Per-page Markdown
Markdown / PlaintextNative (no parsing needed)Indexed directly

​
Cache Properties

  • Hash-based — Identical files uploaded by different users reuse the same parsed output
  • Idempotent — Re-uploading the same file doesn’t trigger redundant parsing
  • Automatic cleanup — When an artifact is deleted or a project is removed, its cached content is cleaned up

​
Parse Status

Each artifact tracks its parsing state internally:
StatusMeaning
noneNo parsed content exists yet
pendingParse job is queued or in progress
readyParsed content is available for AI features

​
What Uses Parsed Content

Parsed Markdown is consumed by several AI features behind the scenes:
  • Chat view tool — When the agent reads a document in text mode, it uses parsed Markdown for PDFs and DOCX files
  • Semantic search — Parsed content is chunked, embedded, and indexed in the vector database
  • Review runs — The review agent reads parsed content to analyze documents
You never interact with parsed files directly. The document viewer always shows the original source file. Parsing is a backend optimization that makes AI features fast and accurate.

​
Limitations

  • PDF parsing supports files up to 25 MB
  • Very large documents may take a few minutes to parse
  • Complex layouts (multi-column, heavy tables) may have reduced parsing accuracy — use the visual mode in chat for layout-sensitive analysis