User-owned · MCP-native · Portable

Your memory,
portable across every AI.

Every AI tool is building memory — and every one is a silo. Lyx Exocortex is the memory itself: a layer below the apps, on infrastructure you own, exposed through a single MCP endpoint that Claude, Cursor, or any agent can speak. Build it once. Every model inherits the full context.

Temporal knowledge graph Least-privilege by default You can make it forget
01 THE SILO PROBLEM

Switch tools, and your context is stranded.

Claude remembers you inside Claude. ChatGPT inside ChatGPT. Cursor, the same. None of it travels. The gap isn't “give an AI memory” — it's give a person one memory that every AI can use, that they own.

🔒
Claude
Memory locked inside.
🔒
ChatGPT
Memory locked inside.
🔒
Cursor
Memory locked inside.
🔒
Your agents
Starting cold, every time.
02 HOW IT WORKS

One layer below the apps.

Not another app with memory bolted on. The memory itself — sitting beneath the tools, exposed through an open interface any client already speaks.

01

Connect a source

Add sources one at a time — files, mail, your public footprint. You see exactly what's readable, and nothing is grabbed wholesale. Least privilege by default.

02

It builds a graph

A self-maintaining temporal knowledge graph captures not just what's true, but when it became true and what superseded it. Stale facts decay on purpose — so it never rots into noise.

03

Every AI queries it

One MCP endpoint. Claude, Cursor, an agent fleet — they all speak the same interface and inherit your full context. Switch models, lose nothing.

03 WHAT'S INSIDE

A graph that maintains itself.

The substance is a living knowledge graph, not a pile of embedded files — tuned for agents that need a precise fact fast, not ten chunks to skim.

Temporal graph

Knows what's true and when it became true — and what it replaced. History, not just a snapshot.

🍂

Decay by design

Ebbinghaus-style salience fades stale facts. A memory that never forgets slowly fills with garbage — so this one forgets, on purpose.

🔌

MCP-native

A single endpoint every client already speaks. No per-tool integration, no rebuilding context per app.

🛰️

Scheduled ingestion

Subagents scan connected sources, detect what's new or changed, and keep the graph current on a schedule.

🎯

Agent-tuned retrieval

Built to return the relevant thing fast — because the consumers are models that need precise context, not a human skimming results.

🔎

Audit & prune

A clean view of everything the system knows, behind your login. See it, question it, cut it. Trust you can inspect.

04 YOU'RE IN CONTROL

You decide what it knows.

“You decide what it knows” isn't a tagline — it's the architecture. Built for least privilege, not “grab everything.”

Scoped access

Least privilege, always

Connect sources one at a time. See exactly what each can read. Revoke it — or make it forget — on demand. Nothing is taken wholesale.

Zero secrets held

No credential storage. Ever.

We never hold a secret. If access ever needs a key, it stays in your OS keychain — not on our servers. Not in v1, not ever.

Straight talk

Honest about processing

When a frontier model is used, your data is processed under zero data retention and never stored on our servers. No “it never leaves your device” when it does.

EARLY ACCESS

Be first to own your memory.

Lyx Exocortex is in early development. Join the waitlist and we'll reach out as access opens.

Opening your email client.

No backend is connected yet. This opens a prefilled email to Lyx.