mempipe
Sign up

Now in public beta

Turn your PDFs and notes into knowledge your agents can use.

Search across them. Build a lasting Markdown wiki with citations that open the source passage.

Check the evidence yourself. Let your agent reuse and update the same work.

Try Mempipe Add to ChatGPT
  • Free plan to start
  • In-app agent: no key needed
  • Ordinary Markdown, no proprietary format
  • In ChatGPT's plugin directory
Ask the in-app agent, in a thread, about a PDF and to save the answer. It searches the paper and writes wiki/attention-workflow.md with citations; open the page from your files. Watch the real 58-second session ↓

A research question becomes a saved page.

A real example using Attention Is All You Need. Three steps, each one a capture of the running app.

58 seconds: ask a question about a paper, save a cited wiki page, open the citation in the PDF, extend the same page. The stills below are frames of the same session. Select one to see it full size.

  1. Ask a question about the paper.

    We put the PDF in raw/ and asked why the Transformer replaced recurrence with attention. A bare question gets a chat reply, so the prompt on screen also names the file to write, the title, and a word cap. The how-to post carries both prompts verbatim.

    The paper open in Mempipe as extracted Markdown, with a raw folder holding the PDF and a wiki folder beside it. The agent panel on the right carries the question that was asked and the first two tool calls the agent ran to find the file.

  2. Keep the answer as a page.

    The agent wrote wiki/attention-workflow.md and left every existing file untouched. The chips in it are citations; opening one shows the passage behind that sentence in the PDF.

    The saved page wiki/attention-workflow.md open in Mempipe, titled Understanding the Transformer with a Why attention section and citation chips. The agent panel on the right reports that it created the file, and a line below its answer reads “Used 4 sources.”

  3. Ask the next question. Update the same page.

    Then we asked how the model knows word order. The agent reread the saved page and appended a cited section on positional encoding. The earlier explanation stayed word for word the same.

    The same wiki page after a follow-up question. The original Why attention section is unchanged and a new How does it know word order section sits below it with its own citation chips. The agent panel reports that the first five lines are unchanged byte for byte.

Captured in a local build of Mempipe, with the in-app agent on a model of our choosing and the PDF extracted by LlamaParse. The video shortens waiting time. Outputs vary with the model, and a citation that resolves proves the passage exists, not that the sentence summarizes it fairly.

Build a wiki from your own PDFs

Free plan. The in-app agent needs no key.

From a folder of sources to knowledge that stays useful

Leave the next session more than a chat transcript. Your research becomes a named page you can reopen, edit, and build on.

  1. 01

    Find the evidence

    Upload PDFs and notes. Search for a phrase or ask a question across your source folder.

    Ask questions about a PDF →

  2. 02

    Keep the useful work

    Ask the agent to save a brief, report, or wiki page as Markdown, with citations back to the sources.

    Build an LLM wiki →

  3. 03

    Check the source

    Open a citation to inspect the passage behind it. Review important claims and correct the page when needed.

    Open a citation →

  4. 04

    Build on it next time

    Add a new source and ask the agent to update the existing wiki. Pick up from the same pages in your next session.

    Update your wiki →

Citations that open the evidence

Follow a citation from your wiki to the source passage. Mempipe connects extracted text to its place in the document, so you can inspect the evidence behind the work.

Explore source search →

“Mempipe turned a folder of PDFs into a Markdown wiki I can still open anywhere. Click a citation and you land on the paragraph it came from.”

David, Relaye.io
A citation from a wiki page, opened. The passage behind it is highlighted in the extracted text and on its page in the PDF.

Your files stay files

Keep source PDFs beside the wiki. The pages your agent writes are ordinary Markdown you can read and edit yourself.

Search and citations connect that work to its sources. If your research includes CSV data, query its tables with SQL alongside your documents.

Explore the LLM wiki workflow →

“Our research lives in plain Markdown files, not a locked-in notebook. Mempipe searches them, cites the source passage, and lets ChatGPT and Claude Code work from the same files.”

Mr. Chua, Founder, Bespoke.talk

Also in Mempipe

All features →

  • Compare PDF revisions

    Review edits, additions, removals, and inferred moves between two PDFs, with changes highlighted on the original pages.

  • Grep PDFs like source code

    Run exact literal or regex search across your PDFs and scope it with file globs.

  • PDF to Markdown

    Convert PDFs to structured Markdown with a source map back to the original page.

Works with your AI agents

What humans see is what agents use

Use the in-app agent, or connect your own through MCP. Each Space is an MCP endpoint, and both agents work with the files, document search, and source references you can inspect in the workspace.

You choose an external agent’s access when connecting it to a Space. With your approval for write access, it can create and update Markdown pages. Ask it to read the existing wiki before making changes, then review the result.

Your Space’s MCP Endpoint page has setup steps for each of these apps. Any other app that can add a remote MCP server and sign in with OAuth can connect with the URL from that page. Sign-in is a Mempipe sign-in, not a key you paste.

Connect ChatGPT desktop →Connect Codex →Other MCP clients →
Every Space is an MCP endpoint
$ codex mcp add mempipe-your-org-your-space \
    --url https://api.mempipe.ai/v1/@your-org/your-space/mcp
$ codex mcp login mempipe-your-org-your-space
Codex, connected to the same Space over MCP, calls search and read_document. The link in its answer opens the passage it read in Mempipe.

Mempipe Notebook

Use it inside ChatGPT

Mempipe Notebook is in ChatGPT’s plugin directory. It shows your PDFs and Markdown notes beside the chat and lets you choose the files ChatGPT starts from. ChatGPT’s answers can then link to the passages they used.

Your files stay in your Space, so the next chat, the in-app agent, and your other assistants all work from the same sources.

This is the way to add Mempipe to ChatGPT on the web. The ChatGPT desktop link above connects one Space through the desktop app’s MCP settings instead.

Add to ChatGPTWhat the notebook does →
Mempipe Notebook in ChatGPT showing Research folders beside an open wiki page, with Use in this chat selected.

Build your first wiki →

Follow the guide from uploading a PDF to a saved wiki page with citations.

Coming from NotebookLM or Gemini Notebook? →

Compare the workflows for document Q&A and persistent Markdown work.

Build knowledge that comes with receipts.

Bring your sources. Find the evidence. Leave behind something worth keeping.

Try Mempipe Read the how-to

Public beta. Start on Free and upgrade to Starter for more capacity. The in-app agent needs no key.

Talk with the founder for help getting started.