Every AI assistant will need your memory.
You should own it.
Trace turns real conversations — messages today; calls and meetings next — into a private, user-owned memory base, one living profile per person. This page is the investment case; the product speaks for itself on the landing.
Why now
The category just proved itself. The user-owned quadrant is still empty.
Sources: TechCrunch, Bloomberg, SiliconANGLE, Epoch AI, Fellow.ai survey data. Figures indicative.
ChatGPT, Claude, and Gemini all shipped automatic memory in 2025–26 — memory is the retention feature, and demand is validated. But every vendor locks that memory inside its own silo, every notetaker organizes it per meeting, and every personal CRM still asks you to type. The intersection — automatic, per-person, user-owned — is unclaimed:
| AI meeting notetakersGranola, Zoom AI | Personal CRMsClay.earth, Dex | Platform memoryChatGPT, Meta | Trace | |
|---|---|---|---|---|
| Capture | automatic, meetings only | manual entry | automatic | automatic — every channel |
| Organized | per meeting | per person | per vendor | per person |
| Owned by | vendor cloud | vendor cloud | the platform | you — private, exportable |
| Readable by | their app | their app | their model | you + any AI you choose |
The market around that empty quadrant is large and moving: personal CRM alone is estimated near $15B (2025), AI meeting assistants at $1–3.5B growing 25%+ a year, anchored by a $163B CRM market and 100M+ people already living in tools like Notion.
Moat
A switching cost that grows while you sleep.
Memory compounds.
Every day of use deepens a private corpus that exists nowhere else — and that no user will rebuild on a competitor.
Privacy by construction.
Data lives on the user's device — no central database of everyone's private life to breach or subpoena. Incumbents whose business is your data can't copy this without abandoning their model.
Connector economics.
A source-agnostic core means each new channel is a thin adapter. The hard part is built; breadth is cheap from here.
Open by design.
Full export and an open memory API mean zero lock-in — and a real shot at becoming the standard substrate personal AI reads context from.
Business model
Sell software and compute — never data.
Prosumer subscription first, aimed at people whose income is their relationships — founders, investors, recruiters, dealmakers: the crowd that pays $30/month for Superhuman.
Your data stays yours forever; you pay for capture and intelligence — connectors, AI enrichment, the daily loop. Device-first storage keeps the cost base thin: we sell software and compute, not data hosting. The endgame is infrastructure — the permissioned memory layer other AI assistants read from.
“We will never monetize user data. We can't — we don't have it. That's the pitch, not a footnote.”
Traction
Built, not slideware.
The full loop runs end-to-end on real data, every day. Its one user is the hardest kind to fake: the founder, running his entire personal and work communication through it.
The ask
Pre-seed: from proven loop to product.
Raising to ship the app and the next connectors — WhatsApp, calls, meetings — plus zero-setup onboarding and a paid design-partner cohort: the first hundred people who never drop a promise again.
If you believe every AI assistant will need a memory layer — and that users will insist on owning it — let's talk. The demo is the founder's real life.