Join the beta — early access in May 2026. Get on the list →
Private beta — early access in May 2026

The memory layer
your digital world
was missing.

Your decisions, lessons, preferences, contacts — the substance of who you are — sit fragmented across fifty tools you don't own. AI tools forget you between sessions. Notes apps make you write. MemoryPrizm is the personal memory layer that's actually yours: extracted automatically from the AI conversations you're already having, portable across every tool you'll ever use, and yours forever.

Start free See how it works

No spam. We email you once when your invite is ready.

MemoryPrizm interface showing the personal memory layer — typed memories extracted from ChatGPT, Claude, and Cursor, unified into one searchable layer.
The problem

Your memory doesn't actually
live anywhere yet.

Every AI you use forgets you between sessions. Every notes app waits for you to write down what you already know. Every vendor owns the memory they DO keep — until they lose it, deprecate it, or close your account. The substance of who you are digitally is real, but it's homeless.

🧐

Built to forget

Every AI session starts blank. ChatGPT's memory is opaque and only lives inside ChatGPT. The thinking you did yesterday is gone tomorrow — or worse, half-remembered in a way that quietly contradicts itself next month.

🔒

Locked to vendors

Your Claude history dies with your Anthropic account. Your Slack history dies when you change jobs. ChatGPT memory disappears the day OpenAI changes its mind. The substance of you, captured by companies whose business isn't your continuity.

🌤

Smeared across surfaces

A decision in ChatGPT, repeated in Notion, contradicted in email. Your real memory is scattered across tools that don't talk to each other and never will. There's no "you" layer — just shards.

How it works

How your memory layer builds itself.

Connect once. From there, every meaningful turn in your AI conversations becomes part of one memory you own — typed, deduplicated, searchable forever. You only show up when you want to look something up, or to read the Sunday digest of what you actually thought about this week.

Connect your sources

Drop in your ChatGPT, Claude, Cursor, or Gemini export — or stay synced. Today: AI conversation history (where most of your modern thinking actually happens). More sources coming.

We extract typed memories

The extractor reads each turn and pulls out the lessons, decisions, facts, preferences, contacts, and bookmarks worth keeping. Roughly 30 memories per hour of conversation, each linked back to the exact turn it came from.

One searchable memory layer

Ask in plain English. Semantic search returns ranked answers across everything you've ever thought about — unified, deduplicated across tools, and yours to read, edit, or export at any time.

A weekly digest

Every Sunday, an email of what you actually learned this week. Auto-compaction keeps your memory small and signal-rich as it grows, merging related memories so the substance compresses without losing detail.

What gets extracted

Six memory types.
One layer that's yours.

Most of your AI conversations are throwaway. The substance — what to remember — falls into a few clean categories. MemoryPrizm tags each memory by type so you can find what you mean, not just what you wrote.

Lesson

"Postgres jsonb columns can't be indexed by default — use a GIN index on the specific path or queries will full-scan."

extracted from Cursor · Mar 14
Decision

"Going with Stripe Connect over building escrow ourselves — the regulatory load isn't worth the 2.9% margin gain."

extracted from ChatGPT · Apr 02
Fact

"Anna's mom's birthday is November 12. She's allergic to sesame, not peanuts — I had it backwards."

extracted from ChatGPT · Jan 21
Preference

"I write in Linear, not Jira. Don't suggest Jira workflows again."

extracted from Claude · Feb 08
Contact

"Diego — freelance illustrator I worked with on the Q4 brand refresh. Reachable on Are.na, fast turnaround on icon work."

extracted from ChatGPT · Dec 03
Bookmark

"clickhouse.com/docs/en/sql-reference/window-functions — the doc that finally made window functions click."

extracted from Claude · Mar 28
Features

Built like memory should work.

Not a journaling tool. Not a knowledge graph you have to curate. A memory layer that gets smaller, not bigger, as it learns who you are. Every feature exists so you can trust the answers, find the source, and never drown in noise.

🔍

Semantic search across everything

Ask in your own words. "What did I learn about React Server Components this year?" returns ranked memories from every connected source — even if you never used those exact words.

🔗

Linked back to the source

Every memory points to the exact ChatGPT, Claude, or Cursor turn it came from. One click and you're back in the original conversation, in context. Provenance for every line in your memory.

📁

Cross-source dedup

Repeat a thought across ChatGPT and Claude? You get one memory, not two. The same lesson stops cluttering your layer just because you talked about it more than once.

📥

Auto-compaction

Related memories cluster and merge into a single richer memory over time. Typical compaction shrinks redundancy by 40–60% without losing detail. Your memory layer gets denser, not noisier.

📧

Weekly digest

A short email every Sunday with the week's lessons, decisions, and facts. The closest thing you'll get to a journal you didn't have to write.

🎯

Lens agents (Pro+)

Spin up separate layers for separate identities — "Work Me", "Side Project Me", "Parent Me" — each with its own sources and digest. Stop bleeding contexts.

Compared to

Notes apps want you to write.
AI vendors want to own you.
We do neither.

Notion, Obsidian, and Apple Notes are blank canvases that punish you for not journaling. ChatGPT's memory only sees ChatGPT, lives inside ChatGPT, and dies with ChatGPT. MemoryPrizm is the opposite of both — a memory layer that builds itself from work you already did, owned by you, portable across every tool.

Capability MemoryPrizm Notion Obsidian Mem.ai Reflect Apple Notes ChatGPT Memory
Builds itself from AI chats Yes No No No No No ChatGPT only
Cross-tool (ChatGPT + Claude + Cursor) Yes No No No No No No
Typed memories (lesson / decision / fact) Yes No No No No No No
Semantic search Yes Limited Plugin Yes Yes No Opaque
Auto-compaction (dedup & merge) Yes No No No No No No
Linked back to source conversation Yes n/a n/a n/a n/a n/a No
Weekly digest of what you learned Yes No No Surface No No No
Export your data anytime Yes Yes Yes Yes Yes Limited No
What changes

What becomes possible when your memory is finally yours.

Builders & founders

The decisions you forgot you made

Six months from now, you won't remember why you picked Postgres over Mongo. Your memory will. Every architectural call, every tradeoff, every "let's revisit this" is captured and findable — including the reasoning that's now stale.

Engineers

A real changelog of what you've learned

Cursor and Claude debugging sessions become a personal lessons-learned layer. The bug you fixed in February is one search away in October, with the exact stack trace and reasoning you talked through.

Knowledge workers

The notes you never had to write

You used ChatGPT to think through a strategy, draft an email, work out a tax question. Months later, that thinking is still there — typed, dated, findable — without you ever opening a notes app.

Curious humans

A year-end "what did I learn?" that's actually true

By December, you'll have a real answer to "what did I learn this year?" — not a vibe, not a guess, but a typed timeline of your own thinking. The most honest year-in-review you'll ever read.

Privacy & trust

Your memory. Your data. Your control.

A memory layer that lives inside someone else's servers isn't really yours. Your AI conversation history is some of the most personal data you have — we treat it accordingly. Encrypted, isolated, never trained on, and exportable at any time.

🔒

Encrypted at rest

Your conversations and extracted memories are encrypted on disk with per-account keys.

🚫

Never trained on

Your data is never used to train any model — ours or anyone else's. Extraction runs against frozen models.

🗑

One-click delete

Wipe your account from a single screen. Memories, raw imports, embeddings, all of it — gone in under a minute.

📝

SOC 2 in progress

SOC 2 Type II audit is planned for late 2026. We won't claim a badge we haven't earned.

Pricing

Free to try. $12/mo if it earns its keep.

No usage gotchas. No surprise overages. Cancel any time and your data leaves with you.

Free

See if it clicks.
$0
forever
  • Up to 100 memories
  • One connected source (your choice)
  • Semantic search
  • JSON / Markdown export
Start free

Pro

Your full memory layer.
$12
per month, billed monthly
  • Unlimited memories
  • All sources (ChatGPT, Claude, Cursor, Gemini)
  • Semantic search & auto-compaction
  • Weekly digest email
  • Linked-back-to-source for every memory
Start free, upgrade later

Pro+

Multiple brains, longer history.
$29
per month, billed monthly
  • Everything in Pro
  • Priority extraction queue
  • Longer history retention (5+ years)
  • Multiple identities / lens agents
  • Earliest access to new sources
Go Pro+
FAQ

Honest answers.

What do you mean by "personal memory layer"?

A single place where the substance of you — decisions, lessons, preferences, contacts, things you've learned — lives across every tool you use. Today, each tool keeps its own slice and none of them are yours. MemoryPrizm pulls the substance out, unifies it, and stores it in a layer that you own, that's portable, and that grows across vendors instead of dying with them. It's the foundation layer your digital world was missing.

Why start with AI conversations?

Because that's where most modern thinking actually happens. You debug in Cursor, strategize in ChatGPT, draft in Claude, troubleshoot in Gemini — and almost none of it gets saved as a real memory. AI chat history is the highest-density input we can extract from today, with the lowest effort from you. Other inputs (email, voice notes, browsing context) are on the roadmap.

Is my data private?

Yes. Your conversations and memories are encrypted at rest, never used to train any model, and you can delete everything from a single screen. We treat your AI history like your email — your data, your control. See the security page for full detail.

What if I delete my ChatGPT account later?

Your MemoryPrizm layer is independent of the source. Once we've ingested an export, the extracted memories live in your MemoryPrizm account regardless of what happens to the original ChatGPT, Claude, or Cursor account. The link back to the source conversation will gracefully degrade, but the memory itself stays. That's the whole point — your memory doesn't die when a vendor does.

Can I export my memories?

Always. Export to JSON, Markdown, or a single zip at any time. No lock-in. Your memories are yours, and they leave with you. Pro and Pro+ include scheduled exports too — auto-zipped to email or Dropbox on your cadence.

How is this different from ChatGPT's built-in memory?

ChatGPT's memory only sees ChatGPT, only lives inside ChatGPT, and is opaque about what it remembers. MemoryPrizm unifies ChatGPT, Claude, Cursor, and Gemini into one memory layer you actually own — typed, dated, deduplicated, linked back to the exact conversation it came from, and readable, editable, and deletable by you.

What does it actually cost?

Free for 100 memories from one source. Pro is $12/month for unlimited memories, all sources, and the weekly digest. Pro+ is $29/month for priority extraction, longer history retention, and multiple identities. No usage overages, no surprise bills.

Does it work with team or shared AI accounts?

MemoryPrizm is built for individuals first. If you connect a shared account, we'll only extract memories from turns that look like yours. Team-aware features (shared brains, scoped sharing) are on the roadmap, not yet shipped.

What sources can I connect?

At launch: ChatGPT exports, Claude.ai exports, Cursor chat history, and Gemini exports. We're adding Perplexity and Poe next. You can also drop in plain Markdown notes and the same extraction runs over them.

Is there a mobile app?

MemoryPrizm is web-first and works great on mobile browsers. A dedicated iOS and Android app is on the roadmap for later in 2026 — we want the web product to be excellent before we split focus.

Your memory has been homeless since the day you got online.
Let's give it a place to live.

Free to start. Three minutes to connect your first source. From there it runs on its own, growing into the personal memory layer you should have had all along.

Start free Join the beta

Built by a small team in Los Angeles. We made MemoryPrizm because we lived inside ChatGPT and Claude for two years and realized we'd forgotten almost all of it — and that nobody else was going to give us our memory back.