Spark
One role, one focus.
$10 / month
- 1 brain
- Core connectors
- Nightly sync
- Claude + ChatGPT access
A memory layer for Claude and ChatGPT
Give your AI a memory.
hippocampOS keeps a private, persistent record of your working life, partitioned into separate brains for each role you play. Claude and ChatGPT call it on demand. You stop re-explaining yourself.
Enter your email address to request access.
Plate I
From a Claude chat● connected
Invoke my work brain. Draft a reply to Priya about the vendor renewal.
work brain · context injected · 0.2s
Pulling from 14 related threads. The renewal is the Meridian contract, $48k/yr, and Priya flagged pricing concerns on May 28. Here's a draft that addresses her two open questions…
Plate I. One sentence in Claude. Your full working context arrives before you finish reading this caption.
Encrypted at rest and in transitWe never train on your dataWorks with free-tier Claude and ChatGPT
Why now
The smartest models on earth still greet you like a stranger.
Everyone senses it. The internet is full of weekend projects wiring up personal digital brains with markdown vaults, local databases, and duct tape. They work, if maintaining infrastructure is your hobby.
hippocampOS is the version you just sign into. A memory that builds itself from your email, your meetings, and your notes, then answers to your AI tools in plain language.
The condition
Context amnesia, observed in three subjects.
Case 01
The re-explaining professional
Every chat opens with your role, your team, your quarter. The model never holds it. You have typed the same paragraph forty times this month.
Case 02
The student
Each session you re-upload the syllabus, the readings, the half-finished thesis. The AI starts blank. So do its answers.
Case 03
The multi-role operator
A day job, a research seat, a side project. Three contexts, none of them stick. You are the index, and indexing is unpaid work.
How it works
From zero to a working memory in one sitting.
STEP 01
Build your brains
Create a brain for each role of your life. Work. Research. The side project. Home. Name it, describe it in a line, done.
Your AI's memory of work should never leak into its memory of home. Partition by role, switch in a sentence.
STEP 02
Feed them what matters
Connect Gmail with smart filters that skip the newsletters. Pull in Granola transcripts, Obsidian vaults, files, or paste anything. You approve every source.
Or take the shortcut. Paste our Brain Seed prompt into Claude or ChatGPT, let it summarize what it already knows about you, and upload the result. Five minutes to a working brain.
STEP 03
Invoke from your AI
In Claude, say "invoke my work brain." In ChatGPT, call hippocampOS from your connected apps. The right context arrives in well under a second. Then you work.
Memory updates both ways: drop a meeting transcript in after a call, and your brain knows it by the next question. Nightly sync handles the rest while you sleep.
What's inside
Built like an instrument, not a black box.
¶ 1
Many brains, not one profile
Most memory features build one blurry picture of you. hippocampOS keeps clean partitions: your work brain answers work questions, your personal brain stays out of it.
¶ 2
Master Brain
One view across every brain you own. Ask what matters across your whole life this week and get an answer that actually spans it. On the Cortex plan.
¶ 3
Brain Seed
A copy-paste prompt that turns your existing Claude or ChatGPT history into a working brain in minutes. The fastest cold start in the category.
¶ 4
Connectors that matter
Gmail, Granola, meeting transcripts from any tool, file uploads, markdown vaults. Notion & Evernote next
¶ 5
A memory you can see
Traverse your memory as a living graph. Every node inspectable, every connection visible. If your brain knows something, you can see why.
¶ 6
Every access, on the record
A full audit log of what was retrieved, when, and by which tool. Your memory keeps receipts.
Where it lives
Works where you already work.
No new chat app to adopt. hippocampOS plugs into the tools you open every morning.
Claude · Native connector
Invoke from any Claude chat
Add hippocampOS once in Claude's connector settings. From then on, any brain answers by name, in any conversation, on any Claude plan. No Max subscription required.
1. Settings → Connectors → Add hippocampOS
2. Sign in once
3."Invoke my work brain"
ChatGPT · Connected app
Call it from ChatGPT
Connect hippocampOS from ChatGPT's app settings and sign in once. Authentication is handled for you. Setup takes about three minutes.
1. Settings → Apps → Connect hippocampOS
2. Approve access
3. Ask, with your memory attached
Evidence
Measured, not promised.
Memory systems get judged on one question: when asked about a life, do they recall it correctly? Here is where we stand on the standard long-term conversational memory benchmark.
| System | F1 score | BLEU-1 | Recall@10 |
|---|---|---|---|
| No memoryzero-shot baseline | 15 – 20% | 10 – 15% | — |
| Published referenceLoCoMo benchmark, Maharana et al., 2024 | ~65.3% | ~58.7% | ~86.7% |
| hippocampOSgeneric paged mode | 65 – 72% | 60 – 68% | 94.86% global · 96.5% scoped |
F1 score, in plain terms
When asked about your life, how much of what it recalls is right, and how much of what's right it manages to recall. Both at once.
BLEU-1, in plain terms
How closely its answers match what a careful, correct human answer would say. Higher means fewer made-up details.
Recall@10, in plain terms
When it searches your memory, how often the right item shows up in the first ten results. Ours found it more than 19 times out of 20.
~1 ms
To select and address the right brain. Choosing where to look costs effectively nothing.
<1 sec
For your context to arrive inside Claude or ChatGPT after you invoke a brain.
~30 sec
For a full recalled answer that takes 2+ minutes when the AI digs through your inbox directly.
Figures from preliminary internal evaluation runs, June 2026. Full methodology and reproducible benchmark results will be published at launch.
Your brain, your data
Trust is not a footnote.
You are handing this product your working life. That deserves more than a paragraph in a privacy policy.
Encrypted, always
At rest and in transit. Not a tier, not an add-on. The default and only mode.
Never trained on
Your data builds your brain. It never trains a model, ours or anyone else's.
Query-scoped retrieval
Your AI receives only the slice of memory your question needs. Never the whole archive.
You control ingestion
Nothing enters a brain without your say-so, source by source, with filters you set.
Delete means delete
Remove a source, a brain, or your whole account. Export everything as markdown first if you like. Gone is gone.
No ads, no brokers
We sell subscriptions, not you. That is the whole business model.
Pricing
Simple tiers. Priced like a habit, not a license.
Synapse
Work and life, kept apart.
$20 / month
- 2 brains
- All launch connectors
- Nightly + on-demand sync
- Email support
Cortex
For the true multi-hat operator.
$30 / month
- 5 brains
- Master Brain, across all of them
- Real-time sync
- Early access to new connectors
Launch pricing, held for early members. Fair-use storage and standard connector API limits apply. There is no free tier; there is a guided demo brain you can explore before paying a cent.
Questions, anticipated
The four things everyone asks.
Do I need a paid Claude or ChatGPT plan?
No. hippocampOS does its own retrieval before your AI ever sees the question, so even free-tier Claude works. You pay your AI provider for intelligence. You pay us for memory.
Is my data used to train models?
No. Not ours, not anyone's. Your data is indexed for one purpose only: so your brains can answer you. That commitment is in our terms, not just our marketing.
What happens if I cancel?
Export everything your brains hold as markdown, then delete it all. We keep nothing you have not chosen to keep. Leaving should be as clean as joining.
When do I get access?
We are onboarding in waves to keep the experience fast for everyone. Join the list and we will write to you when your wave opens.
Begin
Your AI is brilliant. Let it remember you.
Join the waitlist and we will write to you when your wave opens.
Request access