ZABRAIN

Pre-Deck Data Room

Confidential  ·  Pre-Deck Data Room  ·  Do Not Distribute
Zabrain — The first interface designed to stay with you for life.
01 The Category Claim Memory OS. We name it, we own it.

"ChatGPT knows the internet. Zabrain knows you."

Zabrain is the Memory OS. The platform layer that was never built into the computer. Like macOS underneath every Mac app, Zabrain runs underneath every AI tool, every agent, and every workflow. The substrate captures your decisions, meetings, voice notes, and patterns automatically, with no manual input, and builds a structured understanding of you that compounds in value every day you use it.

The consumer product is a cloud desktop. Open it and your entire life connects: Fitness, Finance, Health, Work, Relationships, Food, Travel, Mind, Family, Education, Therapy. Seventeen sections, one graph underneath. Press Cmd+K from anywhere and the Brain Bar opens. Natural language dispatches to any section, any agent, any integration. This is the operating interface. The substrate is what powers it.

Every AI you plug into the substrate. Claude, ChatGPT, Cursor, any MCP-compatible model, reads from the same source of truth about you. You stop re-explaining yourself. The AI stops starting from zero. Cross-platform and user-owned by design; structurally impossible for any of the giants to replicate without breaking their own business model.

Four Surfaces. One Architecture.

SurfaceWhat It Does
Personal Brain The consumer cloud desktop. Brain Bar, 17+ life sections, bento widget canvas. Your private knowledge graph underneath, everything you know, structured forever.
Company Brain Institutional memory that survives headcount changes. When someone leaves, their knowledge stays. Same substrate, org-level access controls.
Expert Marketplace Accumulated judgment runs autonomously. Revenue while you sleep. No time cap.
100-Year Memory Cognition survives the body. The first time this has ever been possible.

All four run on one substrate. The substrate is the defensibility. Each surface requires the one before it. The architecture compounds across all of them.

02 The Problem Three stacked architecture gaps, not product gaps
01
Personal. Every computer ever built starts with zero memory of you. You open ChatGPT and it doesn't know your name. You open Cursor and it doesn't know your codebase. You open any app and it doesn't know what you decided last week, what you built last year, or who you are. You re-explain yourself every single session.
02
Organizational. Someone senior leaves, and everything they knew leaves with them. New people spend months rediscovering what their predecessors already figured out. No enterprise knowledge management tool has ever solved this. They all require human labor to maintain, which means they rot the moment people stop updating them.
03
Civilizational. People die. Everything they accumulated, the judgment, the patterns, the earned wisdom, evaporates. The first generation of humanity that could preserve cognition beyond the body is alive right now. No one has built the infrastructure.
This isn't a product gap. Memory was never built into the operating system.

Why Every Existing Solution Fails

Notes apps store; they don't learn. Wikis rot the moment someone stops updating them. RAG systems retrieve documents but don't build understanding. AI assistants reset every session. Memory inside ChatGPT or Claude is locked to one ecosystem. The failure mode is architectural. You can't solve a memory problem with a storage product. You need a system that maintains itself: distills, connects, verifies, and promotes knowledge automatically, without ceremony.

03 Why Now Five convergent signals. 2026 is the window.
01
AI Agent Proliferation. Every major AI company is shipping autonomous agents. Every agent is stateless; each session starts cold. Without a persistent memory layer, agent workflows break on context. The model layer is commoditizing. Reasoning is becoming a utility. The harness layer (memory, context, history, provenance) is what compounds. Every better model makes the memory substrate more valuable, not less.
02
$185M+ Deployed Into the Category. In 36 months, institutional capital has confirmed this is a real category before anyone's built the product at scale.
03
Limitless → Meta. December 2025. Limitless raised $33M from a16z, NEA, First Round, and Sam Altman. Reached $2M ARR. Acquired by Meta and absorbed into Reality Labs / AR glasses. 80,000 users abandoned overnight. The demand is proven. The supply is homeless.
04
The Golden Cage Problem. Apple, Google, Microsoft, OpenAI, and Anthropic are all building memory, locked to their ecosystem. Your ChatGPT memory doesn't talk to your Apple Intelligence. Memory siloed inside platforms is a golden cage. A cross-platform, user-owned memory layer is structurally impossible for any of the giants to offer without breaking their own business model. We can.

Capital Deployed. Category Validated

CompanyRaisedWho Backed It
Limitless$33M → acquired by Metaa16z, NEA, First Round, Sam Altman
Mem0$24MPeak XV, Basis Set, YC S24
Letta (MemGPT)~$14MUC Berkeley origin
Supermemory$3MSeries A
BlockBrain€23MEuropean institutional
ZepUndisclosedYC + Root + Engineering Capital
Total deployed$185M+Category confirmed at institutional scale
04 The Architecture Three-layer retrieval, five-stage pipeline, zero-trust RBAC

Everything runs on one substrate. The substrate is the defensibility. Most competitors use one storage layer. We built three, composited, because they answer fundamentally different questions.

Three-Layer Retrieval

LayerTechnologyAnswers
Knowledge GraphNeo4j, typed relationships, time-weighted traversalWhat is this related to?
Semantic Searchpgvector, embedding-based similarityWhat is this similar to?
Lexical SearchFull-text, keyword-based retrievalWhat was the exact record?

Five-Stage Intake Pipeline

Nothing promotes to the canonical graph automatically. Every fact passes through five stages. Every node has a traceable history, can be rolled back, and is actively maintained by agents running in the background. The brain never silently overwrites old truths.

01 · Intake
02 · Distill
03 · Connect
04 · Verify
05 · Promote

22 MCP Tools. Live

The brain exposes 22 tools via the Model Context Protocol. Any MCP-compatible AI (Claude, ChatGPT, Cursor, any agent) plugs in and uses the brain as its source of truth. Key tools: orient_brain (full context snapshot of everything relevant now), search_brain (semantic + graph traversal), capture_knowledge (write to graph), absorb_content (long-form → structured nodes).

Zero-Trust RBAC

Information that a connection doesn't have access to literally doesn't exist to that connection. Not filtered, nonexistent at the query layer. This is what makes Company Brain viable for enterprise: an employee can't leak what the graph never shows them. Personal and company memory boundaries are cryptographically enforced.

Consumer Layer

The substrate is what powers it. The consumer interface is what users open. A cloud desktop: bento canvas of widgets, each connected to the memory engine in real time. Press Cmd+K anywhere to open the Brain Bar. Natural language dispatches to any life section, any agent, any integration. Seventeen life sections organized into one view: Fitness, Finance, Health, Work, Relationships, Food, Travel, Mind, Family, Education, Therapy and more. Developers build Brain Apps on the same substrate via the brain.js SDK. Every app on the platform compounds the graph for every user on it.

What's Live Today

ComponentStatus
Memory engine (graph + vector + pipeline)Live, running daily
MCP server. 22 tools exposedLive
Zero-trust RBACLive, enforced at tool-call layer
Git provenance on all canonical nodesLive, full rollback
Internal Company Brain (Edan Labs)Running, all decisions in graph
OS shell. 148 routesBuilt. wiring to memory engine now
Brain Bar + widget canvasSprint 1. active build
05 Traction Engine live, external client onboarded, prior execution credibility
400+
Meetings Indexed
126
Canonical Knowledge Nodes
22
MCP Tools Live
<20min
External Client Onboard Time
Memory engine is live and running daily. The graph, vector search, and five-stage pipeline are in production, used internally for every product decision, investor conversation, and team context at Edan Labs. This isn't a demo environment.
Edan Labs is the internal Company Brain. Every architectural decision, investor conversation, and team context lives in the graph. The company runs on its own product. This is customer zero, and the most rigorous stress test possible.
First external Company Brain client onboarded in under 20 minutes. Tenant created, data ingested, team access configured, live queries working. No white-glove setup. No custom engineering.
Prior execution credibility. This team has raised capital, shipped revenue-generating products, and maintains 100,000+ lines of production code across multiple deployed systems running today. They're builders who raised, not raisers who build.

Phase 0 Gate

The current sprint has one pass condition: 10 named users, 7 days. 8 of 10 send an unsolicited message (text, voice note, email) saying some version of "how did it know that?" No coaching. No prompting. Binary pass/fail. This is the only signal that can't be gamed. Everything downstream (MRR, retention, growth) is downstream of this moment.

06 Market & Competitive Position Clear category map, structural advantages the giants can't replicate

Category Map

CategoryWhoTheir Ceiling
Developer InfrastructureMem0, Zep, LettaNo consumer product, no lifecycle, no provenance. Different buyer entirely.
Enterprise Document RAGBlockBrainStatic retrieval. No cross-session learning. Wrong category.
Consumer Personal AssistantOpenHuman, SupermemoryFlat storage architecture. Can't scale to Company Brain or marketplace.
GiantsApple, Google, OpenAI, MicrosoftLocked to one ecosystem. Golden cage. Structurally cannot go cross-platform.
Memory OSZabrainThe substrate beneath all apps. This is us.

vs Mem0. Most Common Comparison

Mem0 (YC S24, $24M) builds memory for AI apps, a developer SDK. Four-line integration, 90,000 developers using it. Mem0 builds memory for AI apps. Zabrain builds memory for the human coordinating across all their AI tools. Different buyer (developer vs. individual), different scope (one app vs. all apps), different defensibility (four-line SDK vs. compounding knowledge graph with lifecycle). Mem0 → tool. Zabrain → OS. We're not competing for the same customer.

vs The Giants. Why They Can't Win This

Every giant is building memory locked to their ecosystem. This is structurally required by their business model; they need you inside their platform. A user's Zabrain brain works with Claude, ChatGPT, Cursor, and every MCP-compatible tool simultaneously. The giants cannot offer this and maintain their business model. We can. Every time a user has to choose between their preferred AI tool and their memory, we win.

Enterprise Problem in Numbers

Data PointSource
$31.5B lost annually by Fortune 500s to knowledge management failureIndustry research
254 average enterprise apps per organization, all siloedIndustry research
2.5 hours lost per day per knowledge worker on information already in the orgIDC
Replacing a senior knowledge worker costs 50–200% of annual salaryMIT Sloan

The Moat Has a Timer

The brain compounds with time. A user at 30 days has a useful tool. A user at 3 years has an irreplaceable asset: years of decisions, voice notes, meetings, patterns, all structured and retrievable. No competitor can retroactively give a new user 3 years of accumulated memory. Switching cost grows non-linearly. That's not a feature. That's a moat with a timer, and the timer is running now.

07 Business Model Consumer SaaS → B2B PLG → platform take rate → LTV across a lifetime

Personal Brain. Consumer SaaS

TierPricePurpose
Free$01K nodes, 3 capture sources, habit formation and virality
Pro$25/monthUnlimited graph, all sources, primary paying tier
Power$250/monthDeep integrations, priority compute, API access

Company Brain. B2B SaaS

PLG motion: Personal Brain power users become internal champions who pull Company Brain into their organizations. No cold outbound. Bottom-up by design.

TierPriceNotes
Team~$50/seat/monthSmall teams, SMBs. entry point
Business$99/seat/monthPrimary B2B revenue tier
Enterprise$50K–$200K ACVCustom contracts. SOC 2 in progress, required to unlock Fortune 500 motion

Expert Marketplace. Platform Take Rate (Year 3+)

An expert's accumulated judgment runs autonomously. Zero marginal cost once the substrate is validated. Revenue scales with usage, not with expert hours.

ModeTake Rate
Live session (expert present, brain enriches context)10%
Async (brain drafts in expert's voice, expert reviews)15%
Brain-only (expert's brain responds independently)20%

100-Year Memory. Annual Subscription (Year 4+)

Premium positioning. Annual subscription, no direct competition. Pricing power is high; there is no alternative. The estate planning market ($60B+) creates pricing floor analogies.

The LTV Arc. No Competitor Has This

A brain user at Month 1 is paying $25/month. At Year 3–5, they're monetizing via the expert marketplace. At some point, they configure 100-year memory. LTV of a serious user extends across their lifetime. No other productivity product monetizes across a human life arc. The switching cost after Year 3 exceeds any price point a competitor can undercut.

Revenue Targets

MilestoneARR TargetDriver
Month 5$50KPersonal Brain public launch. If not hit: diagnose funnel before any further spend.
Month 8$500KFirst enterprise Company Brain contract. SOC 2 complete. YC cohort onboarding.
Year 1$2MZero paid acquisition. Company Brain pilots beginning.
Year 3$10M+Company Brain B2B compounding. Expert marketplace launches.
08 The Team Vision · Substrate · Distribution, all three present on day one

The three functions that typically take a startup three years to assemble. product vision, technical substrate, and Day 1 distribution, are present from the start. Co-founder relationships predate the company by years. Trust was built through execution before equity was formalized.

Daniel Perez
Chief Executive Officer
Started building software at 15 in Miami. At 16 he co-founded PartyMoves, an event discovery and ticketing app covering 12+ venues across the city that reached 500+ users and sold out multiple events. Same year, from his high school lunch table, he ran a coordinated Supreme drop operation: assigned targets, staged checkouts, Zelle payouts for hits, nothing for misses. He turned a weekly retail lottery into a repeatable pipeline and made tens of thousands as a teenager. He has always seen systems where other people see randomness.

What he has shipped since: Vinculum, a professional marketplace whose architecture became the standard substrate every Edan Labs product inherits. Learn The Markets, a trading education business that generated $5.2M in tracked student profits. Veritrade, a live broker-connected verification platform he built v1 of in three weeks while bedridden recovering from surgery and raised the first check before he was back on his feet. BankForBots, the first bank account built for AI agents, real virtual Visa cards via Stripe with isolated spending identities and per-transaction limits, built in six to seven hours when the team needed it. CreatorOS, a seven-agent content operations system deployed live with creator agencies. And now Zabrain.

The conviction behind Zabrain is not new. In 2019 he built Memoriam, an early substrate for preserving cognition beyond a lifetime: what happens to a person's accumulated knowledge, reasoning patterns, and judgment when they die. The stack couldn't support what he envisioned. He filed it away. Then day two of ChatGPT's public launch, November 2022, the thought came: "The most powerful AI ever built has zero memory of me. Five years from now, it won't know what we talked about today." He started building immediately and has not stopped. Memoriam became one of four surfaces. The original question became the company.

This is not a trend he spotted. It is a problem he has held for years, quietly, while building every adjacent system that would eventually make it possible.

Ships production code on Zabrain daily. He is not a CEO who manages engineers. He is an engineer who runs the company.
Alfred Brewer
Chief Technology Officer  ·  "Fredo"
Alfred's older brother runs in the same circles as Daniel's friend group. When Daniel and Antonio started building Veritrade, they asked the network who had actually shipped something. Alfred's name came up. Daniel showed him the MVP. Alfred saw what it could become and wanted in. They agreed to three months. Within weeks it was obvious the fit was right — for both sides. That was over a year ago. The vision, the pace, and the drive to build remain aligned.

He took Veritrade from a bare-bones MVP built on early frontier AI to a fully deployed Beta v2.2 with a team of five developers, real engineering workflows, and infrastructure ready for scale. Along the way Daniel surfaced a need the whole company felt: memory persistence and synchronization across the entire team. Alfred built it from scratch. The knowledge graph, the retrieval architecture, the intake pipeline, the access control layer. All of it. He became a permanent co-founder the way the real ones do — by solving the hardest problems before being asked to.

What he built on Brain: A three-layer retrieval system composited together — semantic vector search for meaning, lexical search for keywords, and a knowledge graph for relationships. Most products use one layer. Alfred built all three. The intake pipeline doesn't dump raw data into a database — every piece of information passes through an agentic filter that reads what's already in the brain, makes sense of new intake in context, creates connections, and breaks old ones as needed. The access control architecture makes information a connection doesn't have access to literally nonexistent at the query layer. Not filtered — nonexistent. That is the enterprise primitive.

What he built on Veritrade: The entire platform backend — from simple user interactions to a metric calculation engine that orchestrates five-plus data feeds and tackles industry-level data handling challenges, turning a user's click into their public profile without them bearing any of the complexity. Live broker connections across 15-plus institutions.

University of Florida, third year, studying abroad in Switzerland. Building full-time. Every major technical decision since day one. He didn't get hired to maintain a product. He showed up and built the thing that made everything else possible.
Antonio Hernaiz
Chief Operating Officer. "Toto"
Found Daniel on Snapchat, reached out, and never left. A scheduled meeting ran 6.5 hours. That became a working relationship, the relationship became a co-founding dynamic, and the dynamic was stress-tested through years of shared execution before equity was ever discussed. Daniel gave Antonio 50% of the company on trust. No contract. That is the signal.

He is the operational backbone. While Daniel and Alfred build, Antonio runs everything that moves the company forward: GTM, distribution, partnerships, and the community relationships that give every Zabrain product a warm launch. He identifies what needs to happen next and does it without being asked. The company does not stall when the founders are heads down building.

Not technical by design. Every great founding team needs the person who makes the builders unstoppable. Antonio is that person.

Engineering Team

Engineering: Michael Davidson, Riley Lamont, and Isaac Porat. Nick Granberry sits as Head of Sales. Alfred leads all technical architecture and owns the full engineering direction. Georgia Tech is an active sourcing channel for additional engineering hires as the team scales.

09 The Round Pre-seed · SAFE · use of funds breakdown
Pre-seed  ·  SAFE
Round size, valuation cap, and full cap table shared at partner meeting.

Use of Funds

Capital goes to three areas: the people who close the technical gaps fastest, the hardware that unlocks the enterprise deployment model, and the compliance that opens Fortune 500 doors. Nothing else.

Team

RoleWhat They Unlock
ML EngineerRetrieval quality, model fine-tuning, data pipeline hardening from day one
Cybersecurity EngineerEnterprise-grade RBAC, zero-trust infrastructure, breach prevention at the architecture layer
DevOps EngineerScalable deployment for both SaaS and self-hosted enterprise variants
Head of DesignConsumer interface that non-technical users understand on first sight, no onboarding required

Hardware

ItemWhy It Matters
NVIDIA DGX SparkRuns local chunking, embedding, and reranking models simultaneously. Required for the self-hosted enterprise deployment model.

Compliance

ItemWhy It Matters
SOC 2 Type 1 (~$18–22K)Required gate for Fortune 500 Company Brain contracts. Security infrastructure built in from day one, not retrofitted.

Why This Bet, Why Now

The substrate is built and running. This isn't a thesis on a deck. It's a live memory engine with a knowledge graph, 22 production MCP tools, a deployed Company Brain, and an external client onboarded in under 20 minutes. The risk profile is execution, not architecture.
The category is confirmed at unicorn scale before anyone has shipped it. $185M deployed, Limitless acquired by Meta. We are the team that is actually building.
The moat has a timer. The brain compounds with time. The first company to create the "how did it know?" moment at scale and retain users through Year 1 owns the category. We are at the gate.
Cross-platform, user-owned memory is structurally impossible for the giants to offer. Every time a user wants their memory to follow them across tools, we are the only option. That isn't a niche. That is the internet.