Introducing Memo: a knowledge graph for everything you file

Most of us are better at collecting than organising. Reports, articles, web pages, contracts, meeting notes — they arrive faster than we can ever file them properly. The tools we use to store all this are good at holding documents and bad at understanding them. A folder will keep your PDFs safe forever, but it will never tell you that the organisation named in one report is the same one mentioned in three others. That work — the connecting, the cross-referencing, the remembering — still falls to us, by hand, one document at a time.

Memo is our answer to that. It is a hosted knowledge-graph document platform: you upload what you file, Memo reads it, and it builds a single searchable graph of everything inside. The tagline says it plainly — a knowledge graph for everything you file.

The problem with storage that doesn't read

Storage solves the wrong half of the problem. Knowing where a document lives is easy. Knowing what is in it, and how it relates to everything else you have, is the hard part — and it is the part that quietly never gets done.

So we link by hand. We tag, we rename, we keep a running document of "where things are". It works until the pile grows past the point where any one person can hold it in their head, which happens sooner than you would think. After that, your knowledge is technically saved and practically lost: present somewhere, findable only if you already remember it exists.

What Memo is

Memo is different in one specific way: it reads your documents and connects them automatically, so the relationships are there waiting for you instead of being something you have to build.

You upload PDFs, web pages by URL, markdown, HTML, plain text and notes. Memo turns each one into searchable passages, reads them for the people, organisations, places and concepts they mention, and works out the relationships between those entities. When the same entity appears across several documents, Memo resolves it to one canonical node — so an organisation mentioned in five reports becomes a single point in the graph with five sources behind it, not five disconnected copies.

The result is a graph you can actually use. You can search by meaning rather than exact keywords, and you can ask a question and get an answer drawn from your own corpus, with citations back to the documents the answer came from. Answers can stream as they are composed.

What happens when you upload

It is worth walking through, because the shape of it matters.

When you add a document, the bytes go straight from your browser to secure object storage through a one-time upload URL. Memo's own services coordinate the work but never sit in the byte path — your files take the shortest route to storage. From there Memo:

• Chunks the text into passages — markdown, HTML and text natively, and PDFs via PyMuPDF (with a poppler pdftotext fallback), all UTF-8 sanitised.
• Embeds each passage with an embedding model so it can be searched by meaning rather than by literal words. The same model embeds your queries, so a search and the passages it ranks speak the same language.
• Extracts entities and typed relationships using a language model — the people, organisations, places and concepts, and how they connect.
• Resolves those entities to canonical nodes, so the same thing mentioned in different documents becomes one node.
• Stores the lot in Postgres with pgvector, where reads collapse back down to those canonical nodes.

You do none of this. You upload, and a little later your document has joined the graph.

Who it's for

Memo is for anyone whose work is mostly reading and remembering across a lot of material — researchers piecing together a literature, analysts tracking the same organisations across filings, anyone building a long-lived reference library out of documents that keep arriving. If you have ever wished your files could answer a question instead of just sit there, this is built for you.

It is also for developers. Memo speaks MCP over HTTP, so you can connect your own tools and assistants and query the graph, ask cited questions, and list and read sources from wherever you work. There is an HTTP API and a CLI built from the open docgraph-mcp source as well. See the MCP guide and the API reference when you are ready for that.

Try it

Memo is hosted at www.document-analyser.com and you sign in with Google — there is nothing to install to get started. The Free plan lets you try the whole pipeline with files up to 25 MB each; Pro is $19 a month with a 500 MB per-file limit; and Team is custom for groups that need more.

The honest way to understand Memo is to give it a handful of documents you already know well and then ask it something you would normally have to dig for. Start with the getting started guide, and have a look at pricing when you want to go further. We think you will like watching the graph build.

Made by Gamma Systems Pty Ltd in Australia.