mempipe

Ask questions across every PDF in a Space — and open the evidence

Mempipe's in-app agent answers questions across the PDFs and notes in a Space, using the passages it retrieves as evidence. A valid citation opens its source passage. Save useful answers as Markdown notes that you and your agents can build on.

Start on Free. The in-app agent uses your own provider API key, and that provider bills you for model use. See how to set it up.

One question, your whole corpus

Real questions rarely live in one file. The pricing history is in one report, the churn numbers in another, the caveat in a meeting note.

In Mempipe, your documents live in a Space: a durable, searchable workspace, not a disposable chat context. Ask a question and the agent searches the PDFs and Markdown notes in the Space, reads the passages it finds, and is instructed to answer only from what it read.

Citations that open, not decorate

A citation should take you to the evidence. Mempipe citations resolve through a source map. Each cited passage is anchored to its location in the original document, so a click lands you on the exact passage, highlighted, with its page in the PDF open beside it.

  • →Passage-level citations. A citation points at one passage, not at a whole 40-page document.
  • →Built to decline gracefully. When your documents don’t contain the answer, the design goal is “not enough evidence” — not a fluent guess.
  • →Answers you can keep. Ask the agent to save an answer as a note in the Space, copy it as Markdown, or download the whole chat. The chat itself is gone when you reload the page, so keep what you need. A saved citation keeps opening its passage, though renaming, moving, or changing the cited file can break it.

A workspace, not a chat window

Because the Space is file-native, everything around the Q&A keeps working like files: documents stay organized in folders you control, each PDF keeps its source-mapped Markdown view linked to it, and your notes live alongside the sources they cite. Agents connect over MCP and use the same search, documents, and citation links you use in the workspace, so their evidence comes from the index you search.

Learn how

Frequently asked questions

How many PDFs can I chat with at once?
All of them in the Space you're working in, once Mempipe has processed them. The agent can search and read the whole Space, so you don't need to open a file first. It can also narrow its search to a single file, or to a folder with a glob pattern such as reports/**. How much you can store depends on your plan. Free allows 100 files and folders and 200 MB, shared across your organization's Spaces.
Where do the answers come from?
From your documents, through a model you choose. The in-app agent uses your own API key from OpenAI, Anthropic, Vercel AI Gateway, or OpenRouter, and that provider bills you for model use. Questions retrieve matching passages from the Space's search index, and the model uses those passages to answer. The prompt asks it to cite the evidence and say when the documents are insufficient, but important answers still need review.
Can I verify an answer?
That's the point. Citation chips open the exact passage in the source document, including the page and highlight in the original PDF. Models can still omit or misuse citations, so Mempipe keeps the evidence close enough to audit instead of asking you to trust the answer.
Can my AI agents ask questions too?
Yes. Every Space is readable over MCP, so agents get the same search, the same documents, and the same citations as the workspace UI. A valid citation in an agent's answer opens its passage in Mempipe.

Explore more

Build knowledge that comes with receipts.

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.