Discuss research

NeoLab / Context and memory

Papyrus

Turns documents into Markdown that agents can read and cite, with structure and page anchors preserved and no LLM in the conversion path.

PrototypeStarted 26 Aug 2026Updated
From document to citable contextIllustration, not results

The problem

Agents read documents badly. PDFs, slide decks and spreadsheets lose their structure when flattened to text, and answers built on them cannot point to where a fact came from.

Why it matters

An agent that cannot cite its sources cannot be checked. Provenance starts at ingestion.

Research question

Can deterministic conversion keep enough structure and provenance for agents to cite their sources, without using a model to do it?

Approach

  • Converts PDF, Word, PowerPoint, Excel, HTML, EPUB and more into Markdown, keeping headings, tables and page anchors.
  • No LLM in the conversion path, so the same file always gives the same output.
  • Ships with an MCP server so agents can call it directly.

Methodology

  • The code includes a side-by-side comparison against naive text extraction.

Findings so far

No measurements published yet.

Limitations

  • Conversion quality has not been measured on a fixed document set.

Artifacts