Memory for your AI · imagined 80 years ago
A memory for your AI that comes with the page number.
Memexa is the memory you own: a knowledge graph built from your notes, documents and scans. Claude or ChatGPT read it, cite it down to the page and keep it in order — EU-hosted, exportable forever.
Free plan, free forever · no credit card · cancel anytime
Example scene: a scanned insurance PDF lives in Memexa. Asked about the deductible, the AI answers with the amount and cites the source — document, page 4 — while the passage is highlighted in the scan.
Scripted illustration of real tool calls.
Works with ChatGPT · Claude · Cursor ● EU-hosted · GDPR ● Markdown · export any time
Two gaps, every day
You keep explaining things that are already written down.
Gap 01 · Between your tools
Your AI tools don't talk to each other.
ChatGPT for one thing, Claude for another, Cursor in your code — and each starts from zero. What you taught one is invisible to the next.
Gap 02 · Between AI and your knowledge
Your knowledge sits where no AI can see it.
In PDFs, email threads, scattered notes. It's written down already — no AI can read it there. So you explain it a third time.
One thing closes both gaps: a memory every client can read — and that belongs to you.
The Memexa file
Three things worth remembering. Memexa remembers the rest.
File 01 · Remember
Your AI forgets you. Memexa doesn't.
Notes, documents, decisions — everything becomes a linked knowledge graph your AI searches and cites down to the page.
File 02 · Model-neutral
One memory, every AI.
Claude today, ChatGPT tomorrow, the next model after that. Models come and go — the brain stays.
File 03 · Yours
It belongs to you.
EU-hosted, all Markdown, export any time. And it tends itself instead of keeping you busy.
A look inside the brain
This is what knowledge looks like when it's connected.
Travel tips, lab results, recipes, project decisions, the garden: everything you save becomes a node — and Memexa draws the connections. Your AI follows them when you ask.
● Topic ▭ Note PDF document ◠ Person Example excerpt — your graph builds itself as you save.
The five-minute moment
Connect. Ask. “My AI knows me.”
No migration, no new tool to learn: connect your AI client once, toss in notes and PDFs — and from the first question on, your AI answers from your own material. With the source: document, page.
- 01 Connect — once, five minutes, no technical skills
- 02 Toss it in — notes, PDFs, links, decisions
- 03 Ask — every answer cites its source
Example scene: something is saved in one AI client — Memexa stores and links it via a tool call. Days later the same person asks another client and gets the answer with its source.
Scripted illustration of real tool calls.
An honest comparison
Not built-in memory. Not a context file. Yours.
| Memexa | Obsidian + plugins | Notion (+ AI) | Built-in memory | |
|---|---|---|---|---|
| Where your data lives | EU — Scaleway, Paris | Your device | US | US |
| Open source | No — export is open | App no, files open | No | No |
| Reads scans & PDFs (OCR) | Yes | “clunky” — community's words | Attachments, superficially | No |
| Page-level sources | Yes | No | No | No |
| Any MCP client | Yes — one URL | Via community servers | No | No, tied to the tool |
| Price | Free · from €12/mo | Free, sync extra | from ~€10/mo | Included in AI plan |
As of August 2026 · based on publicly documented behavior
Before you ask
The five questions that always come.
Why not just use Obsidian?
If your knowledge is text you write yourself: use Obsidian, it's excellent at that — and your vault imports later as a ZIP with wikilinks intact. The difference is documents. In the Mac Power Users forum, an Evernote emigrant calls Obsidian “good at text only”, with getting receipts and scans in being “clunky”. That exact part is Memexa's core: scans go through OCR and become citable down to the page.
Where does my data live — and can I leave?
The exit first: everything exports as Markdown, one click, complete — the graph is an index, not a lock-in. And inside: servers, database and documents live with Scaleway in Paris; AI processing runs through Mistral AI, also Paris. An EU company address is not EU hosting — which is why we publish names and places.
What does it cost — and what do you live on?
Free stays free forever; Pro is €12, Power €29 a month. We're bootstrapped, a team of two — no investors, your subscriptions are the entire business model. Which is why Memexa won't go the way of Pocket, Omnivore and Rewind: there's nobody to pull our plug.
Does it fit my stack?
Memexa is an MCP server — the open standard spoken by Claude, ChatGPT, Cursor and practically every new AI client. One URL, one browser sign-in, done. No per-tool plugin, no API keys, and the next client you try can read along immediately.
How do I know it works?
The demo above is scripted — the Free plan isn't. Connecting costs nothing and needs no credit card. The first question you ask about your own uploaded PDF answers this better than any landing page.
Pricing, honestly
Three plans. No asterisks.
Free
€0 forever
—
For trying it out and small graphs.
- ✓ ~150 notes
- ✓ 50 document pages
- ✓ 1 AI client
- ✓ 20 questions per day
- ✓ All features, full export
Pro
€120 per year
works out to €10 per month
Your daily memory, without everyday limits.
- ✓ Unlimited notes
- ✓ 500 document pages / month
- ✓ All AI clients at once
- ✓ 200 questions per day
- ✓ Everything in Free
Power
€290 per year
works out to €24 per month
For heavy document volumes and intensive use.
- ✓ 3,000 document pages / month
- ✓ 1,000 questions per day
- ✓ Priority processing
- ✓ Everything in Pro
Free stays free forever · paid plans cancel monthly · export any time
Verifiable
An EU company address is not EU hosting.
So instead of badges, here are names and places — every claim checkable.
Data center
Scaleway, Paris (fr-par) — servers, database, object storage. Named, not just “EU”.
AI processing
Mistral AI, Paris — OCR, embeddings, distillation. No US model provider sees your content.
The exit
Markdown export, one click, complete — before you have to ask.
Error tracking
Self-hosted (GlitchTip on our own EU infrastructure) — no third-party telemetry.
From the manifesto · 1945
“In 1945, Vannevar Bush described a machine that never existed: the Memex. Eighty years later we have machines that can think. But they don't remember.”
Tell your AI once.
From today on it remembers — in every client, with sources.
Free forever · cancel anytime · export in one click