Why Mempipe is betting on Markdown
Markdown is not just a writing format for Mempipe. It is the open, readable layer where files, citations, wiki links, and agent context can meet.
Open files, useful memory
Notes on the Mempipe bet: Markdown as durable substrate, wiki-style links as workspace structure, and source-mapped files as context that humans and agents can share.
First cluster
Start with the Markdown thesis, then read about LLM wikis, PDF-to-Markdown, and citations.
Markdown is not just a writing format for Mempipe. It is the open, readable layer where files, citations, wiki links, and agent context can meet.

Karpathy's LLM wiki pattern applied to one PDF: the exact prompts, the saved page, the citation that opens the source, and Claude Code plus Obsidian compared.

A sneak peek at the PDF comparison prototype we’re building for Mempipe.

We pit frontier models against recent low-cost, near-frontier models to find the cheapest reliable citation model. We also found that prompt engineering with explicit syntax and a worked example still matters.
The useful path is not PDF extraction by itself. It is source-mapped Markdown that can be searched, cited, and opened back to the original evidence.
Chat history is useful memory for a session. Wiki-shaped files are better memory for work that needs names, links, citations, and reuse.
Bring your sources. Find the evidence. Leave behind something worth keeping.
Public beta. Start on Free and upgrade to Starter for more capacity. Bring your own API key for the in-app agent. Reading, searching, and editing files needs no key.