Confidente

Investor Memo

Jonathan Merlin Greenberg — July 2026

1. TL;DR

Every AI assistant is a brilliant stranger who studies your profile. The one that truly understands you has never existed. Not until now.

Confidente is the first AI assistant built on a Persistent Context Layer (PCL) — infrastructure that maintains a structured, durable, user-owned understanding of who you are, organized into user-shaped Lanes (Personal, Business, and Health as sensible defaults, fully customizable). It runs on frontier models, so the user gives up nothing on raw intelligence. What they gain is the layer no model provides at this depth, and no single lab is structured to provide at all: a coherent, compounding understanding of the person asking. The result is an assistant that doesn't reason from an inferred sketch of you — it reasons against the full picture.

This memo is written to a discipline: every claim in it is true, in its tense, on the day you read it. It lives at this address and is updated, with dated notes, as the product advances — so what you are reading is always the current state of the company, not a snapshot. It makes three kinds of claims, labeled plainly:

The idea. A context layer above the models — structured, governed, portable, and owned by the user — is the most defensible position in consumer AI, and it is a position no lab can occupy. Part One makes that argument against the strongest current versions of what the labs have shipped, by name.

What is already proven. The founder has run his professional and personal life on the working prototype of this architecture, daily, for over a year — and the product built from it is live, access-gated, in daily use at confidente.app. Part Two is an exact inventory: nothing in it is projected.

What is not yet proven. That the effect transfers — that a person who didn't build the system experiences the same transformation the founder does. Part Three names that question honestly, because it is precisely the experiment this raise funds.

Confidente is raising $750k to fund a 15-month sprint: private beta, then web and mobile launch, aggressive customer acquisition during the category-opening window, retention and unit-economics validation, and arrival at the next round with cash still in the bank.

Part One — The Idea

2. The Opportunity

Persistent understanding is the unsolved problem in AI assistants. Memory exists now, and the newest versions of it are genuinely good — what is still missing underneath it is understanding. Anyone who has used Claude or ChatGPT seriously for more than a few weeks has felt the gap directly: what carries forward is a flattened composite — facts the system happened to infer, pooled without partition, held without the reasoning that produced them — while the actual shape of the work stays where it always was, with the user. The products are remarkable; the understanding underneath them is thinner than they are.

Every major lab is now investing in memory at an accelerating pace — and that is the most useful signal in the market. In June 2026, OpenAI rebuilt ChatGPT's memory around background synthesis; the system now maintains a profile of each user automatically and keeps it current. Google has gone furthest: Gemini's Memories learn automatically, its Personal Intelligence reasons across a user's Gmail, Photos, and Search history, and its Daily Brief now assembles a personalized morning digest for every Google AI subscriber — with Gemini heading into Siri, and default distribution on the iPhone with it. The largest AI companies in the world are spending flagship releases teaching hundreds of millions of people to expect an assistant that knows them — validating, with the strongest versions their business models allow, the category Confidente is built to own.

That qualifier is the whole story. What every lab is building is a profile, and a profile is not an understanding. Native memory — even in its newest, synthesized, cross-app form — distills one undifferentiated dossier of preferences and facts, written by silent inference, in service of better replies on that lab's platform. Confidente's PCL is different in kind, and the differences are structural, not features: it is partitioned the way a life is, rather than pooled; its writes are governed rather than inferred; its history is preserved rather than overwritten; and it is owned by the user — built to follow them across models rather than bind them to one. The difference isn't whether it remembers. It's whether it understands — and who it works for.

Why now? Three things are true at once for the first time. Willingness to pay for AI assistance is proven and enormous — tens of millions of people already pay for Claude and ChatGPT. The labs have publicly validated the memory gap by shipping partial answers to it, and are training the market's expectations at their own expense. And frontier models have become good enough that a product built on top of them can deliver best-in-class intelligence while competing on something the models don't provide — understanding. The category is buildable, the demand is demonstrated, and no product yet occupies the position. The first months of that window are when defaults get set.

3. Why the Labs Can't Own This

The obvious question deserves a direct answer, twice over: won't the labs just build this? — and its sharper 2026 form, doesn't Google already do this? Confidente competes with Claude, ChatGPT, and Gemini as consumer apps, so the answer has to hold against the strongest thing each has actually shipped, not a strawman. It does, on five structural axes.

First — and this is the load-bearing one — Confidente is model-agnostic, and the labs cannot be. Each lab's assistant exists to sell its own model; it cannot route to a competitor's frontier model when that model is better for the task. The conflict is architectural, not merely commercial: Confidente's understanding lives in the agent layer, above any single model — the way a coding agent now lets a developer swap the model behind it without losing the thread — whereas a lab's memory lives inside its own product and has nowhere else to go. Confidente is not betting on out-modeling Anthropic, OpenAI, or Google; it competes on the layer above the models, a position no lab can occupy without abandoning its own business. The architecture is valuable precisely because no lab can be its neutral home.

Second, structure versus profile. The labs keep rebuilding memory — 2026 proves it — but what they keep rebuilding is a profile: one synthesized portrait, written by inference, tuned to flavor answers for hundreds of millions of median users. A single pool of memory eventually hits the problem every serious user recognizes as context bleed: a health detail coloring a work answer, a personal worry surfacing in a professional draft. Watch what each architecture does when it confronts that problem. OpenAI's answer, in ChatGPT, is a wall — a project either sees the entire profile or is sealed off from all of it, all or nothing. Google's answer is more capture, not more shape: one profile, fed by more of your apps. Confidente's answer is a structure. Lanes keep contexts separate by default, the way a thoughtful person keeps them separate, so the right one is brought to each conversation and nothing bleeds in uninvited. A wall is either up or down; a structure has shape — and shape is what lets a boundary be governed rather than binary. That is not a setting a competitor adds over a weekend: a single profile has no unit to govern, and an all-or-nothing switch is the most that architecture can express. Lane separation is live in the product today; selective, permissioned bridging between Lanes — letting one chosen detail cross when it genuinely should — is the natural next layer, and Confidente owns the structure that can hold it. The labs do not.

Third, the record. Every lab's memory silently overwrites itself: yesterday's understanding is replaced by today's, and the reasoning that led anywhere is gone. Confidente's context discipline preserves it — decisions carry what was chosen, when, and why, and superseded thinking is kept as history rather than erased. This has run daily on the founder's prototype for over a year, and it changes what the assistant is: not a mirror of your current state, but a witness to how you got there. No lab offers this, and none has signaled it — because a profile tuned to flavor the next answer has no use for its own past.

Fourth, ownership and portability. A lab's memory deepens the user's commitment to that lab's assistant, is not built to follow them anywhere else, and dissolves the day they leave. The labs' own "portability" gestures prove the point: Anthropic and Google both now ship import tools that pull your history in from competitors — one-directional funnels pointed at themselves, switching promotions dressed as openness, with no receiving end for anything going the other way. Confidente is built on the opposite commitment, structurally: the user's context is the product's actual working artifact — structured, readable, and designed from the first line to be carried to whichever model is best — and it will never have a second customer of any kind. A lab can match those words; it cannot match the structure, because its memory exists to serve its platform.

The same asymmetry appears one layer down, in whose customer the user is. A Confidente user's account is with Confidente; the models are suppliers. A request carries the conversation and the relevant Lane context and nothing else — no name, no phone number, no email, no login, no account identity of any kind. A lab cannot offer that arrangement, because using a lab's assistant means holding an account with the lab: there, identity and inference are the same relationship by construction. Confidente separates them, and the separation matters — the model sees what you say, and only Confidente knows who you are. Ownership is not only a question of where context can be taken. It is a question of whose customer the user was in the first place.

Fifth, action. Memory that only flavors replies is the smaller half of the opportunity. Google's Daily Brief and OpenAI's Pulse both now assemble proactive daily surfaces — so a morning digest, as a surface, is no longer novel, and this memo doesn't claim otherwise. The difference is what feeds it. Theirs are engagement cards inferred from a profile and scraped from apps: things you might like. Confidente's proactive layer is grounded in governed working state: the decision left open, the commitment coming due, the thread that went quiet — what the user is accountable to, because the system holds the state those obligations live in. One is a feed. The other is a chief of staff.

And the deflationary counter — "won't bigger context windows dissolve all of this?" — has the answer backwards. Raw window size is the labs' territory, but nominal capacity has never been the real limit: models degrade as context fills, because attention dilutes across everything present — a problem Google's own personal-AI work names directly. What determines quality is not how much can fit but what is selected to be there. Selection is exactly the layer a third party can own — and every increase in window size makes disciplined curation more valuable, not less, because it widens the gap between a model handed everything and a model handed the right things. The ceiling was never the real limit; selection is.

The long-run thesis follows. AI assistant relationships compound: the assistant that holds the deepest, most coherent context of a user becomes meaningfully more useful than one that doesn't, and that gap widens every month. Adopting Confidente carries real friction up front — you're changing the assistant you reach for — but that friction inverts into the moat. Once an assistant genuinely understands your life, going back to one that doesn't is a downgrade you feel daily. Switching cost rises with depth. Once a user trusts a single product with the understanding of their life — personal, professional, health, and whatever else matters — they do not casually replace it. We intend to be that product.

4. The Product

The clearest way to understand Confidente is to walk through the experience it is built to deliver. (Part Two draws the exact line between what is live today and what ships next; nothing here blurs it.)

The user opens Confidente, and the assistant already knows them. They don't restate their projects or re-explain which part of their life they're asking about. They ask the question on their mind, and the answer arrives with the full, coherent picture of who is asking — drawn from the right Lane, grounded in history the assistant has synthesized rather than been told again. This is the whole experience: not a chatbot working from what it happened to infer about you, but an assistant that has been paying attention. This core — the chat client, the Lanes, and the persistent understanding injected beneath every conversation — is live and in daily use now.

Lanes are the structural primitive — coherent contexts Confidente keeps separate so it can apply the right one to each conversation. The product ships with three sensible defaults: Personal, Business, and Health. Users rename them, remove them, or add their own as their lives require — a freelancer might split Business into per-client Lanes; a caregiver might add one for a parent's affairs. The Lanes follow the user, not the reverse, and Confidente learns each one organically through use. There is no intake questionnaire; onboarding is frictionless by design.

For Claude and ChatGPT, the chat surface is the product. It carries all their weight — and capable as it is, it's built to be a workhorse for hundreds of millions of people. For Confidente, that same surface is the floor, not the product. The understanding above it is where the value lives, which frees the surface to be what an incumbent can't afford to make it: not just capable, but a pleasure to use.

The launch build completes the experience around that live core, and each piece is design-settled rather than speculative:

The model becomes the user's to direct. Confidente will choose a sensible frontier model by default, with a quiet control to pick the engine behind any conversation — and change it mid-thread without losing a thing. Ask a hard technical question on one model, carry the same conversation to another when the work turns to writing, dial the reasoning deeper when a decision demands it — and Confidente stays Confidente throughout: the same understanding, the same register, a different brain doing the thinking. This is mechanically possible for Confidente and impossible for any lab, for the structural reason in Part One: the understanding was never inside the model to begin with.

The day begins with the Morning Digest — a brief, organized read of what's relevant across the user's Lanes: open tasks surfaced from prior conversations, a follow-up window opening, a health-pattern observation tied to something flagged last week. Short by design; it surfaces what matters, not everything. This digest is not hypothetical — the founder's prototype has generated it every morning for over a year; the product version runs on the same engine.

Notifications arrive when context warrants. Some are table stakes — reminders the user asks for, set invisibly in the flow of conversation. The distinctive ones are the notifications the user never asked for: because synthesis runs continuously across each Lane, Confidente recognizes moments worth surfacing on its own. As Part One argued, the differentiation is not the surface — it's the governed, obligation-grounded state feeding it.

The understanding is inspectable and carryable. Each Lane is a living document the user can open, correct, and export in full — and bringing existing context in is first-class: Confidente is being built to import the structure a user has already accumulated in Claude or ChatGPT projects and synthesize it into Lanes, so the first session opens already informed. The labs offer a readable memory page; the difference is what's on the page — structured working state with its history, not a summary — and whose it is.

Voice carries the same register and the same understanding — a spoken conversation with something that already knows the context, not a smart speaker taking dictation. The voice mode is built and working in development; it deploys behind the access gate ahead of launch.

Because the PCL understands the whole person, Confidente's usefulness is unusually broad — and the breadth is a consequence of the architecture, not a separate set of features. The same assistant that organizes someone's personal life is the one that understands their work, and for a technical user, their codebase and the decisions behind it. A specialized tool can be excellent inside its lane and blind everywhere else; Confidente's advantage is that it isn't specialized — it knows the person, and a person is not a single use case.

5. Position & Voice

Confidente has a distinctive register: substance over flattery. It is built to be supportive the way a good professional is supportive — calibrated, grounded, attentive — rather than the way a friendly companion is. It does not validate for the sake of validation. It tells the user what it actually thinks, with the full context of their situation in mind.

This is a deliberate counter-position. Most consumer AI products are racing in the opposite direction — toward warmer, more agreeable, more validating affect — because users in single-session interactions reward warmth. The trade is that the warmth is context-blind. An assistant working from a profile rather than an understanding can only flatter generically.

The substance of Confidente's voice is downstream of its architecture. Calibrated support requires actually knowing the person being supported — what they're working on, what they've tried, what's going on in adjacent parts of their life. Without context at that depth, every "you're doing great" is shallow praise. With Confidente's PCL underneath, the same encouragement lands on real ground: it can point to specific progress and name a real obstacle, because it actually knows the work.

This is why the voice is defensible. A competitor cannot copy substance over flattery by changing a system prompt; the register depends on the context layer beneath it. The position is a strategic bet on a specific kind of user — people using AI for real work, real decisions, real life. They reward substance and tire of flattery quickly. Confidente is built for them.

6. Market

The serviceable market is straightforward to size: every paid subscriber of an AI assistant has already demonstrated willingness to pay for AI help and would benefit from a PCL underneath it. ChatGPT counts roughly 900 million weekly active users globally — a base that doubled from 400 million in a single year — with more than 50 million paying subscribers and an estimated billion-plus monthly users. Claude serves an estimated 19 million monthly active users on the web alone, with paid subscriptions that more than doubled through early 2026. Combined, the two companies are generating well over $50 billion in annualized revenue and still growing at multiples per year. Confidente does not need a meaningful share of this market to be a substantial company.

The two user pools are differently shaped, which is the strategic crux of this section — and the labs' own usage research confirms it. Anthropic's data shows Claude is used primarily for work: coding, research, technical problem-solving, by developers, researchers, and knowledge workers doing high-context professional work. OpenAI's data shows the opposite skew for ChatGPT — more than 70% of consumer usage is now non-work, up from 53% a year earlier, across a roughly gender-balanced, age-distributed audience increasingly oriented around life management, family decisions, learning, and personal projects. Both pools pay; they reward different things.

This is one product with two messages, not two products. To the professional user, Confidente leads with depth and disciplined context across demanding work — including the codebase, the client, the venture. To the broader audience, it leads with calm, useful daily structure across the parts of life that matter. The product underneath is identical; only the framing changes. The Lane architecture supports both: a developer organizes around projects, a parent around family members, a multi-business operator around ventures. None of these users is an edge case.

Go-to-market leads with the professional and technical user — where the proof is deepest, where the "it already knows my project" demonstration is most concrete and immediate, and where the founder has lived the use most fully. That is the beachhead, not the ceiling. The broader life-management audience is the expansion thesis and the larger long-run market. Which audience converts fastest, retains longest, and supports the highest willingness to pay is genuinely open — and it is precisely the kind of question this raise is structured to answer with first-cohort data, not to pre-decide here.

Part Two — What Is Already Proven

7. Proof

Two things are already true, and both are verifiable rather than projected.

The product is live. Confidente runs in production at confidente.app today, behind an access gate: its own chat client; the Lanes system in full (create, rename, reorder, remove — the user-shaped defaults described above); the persistent context layer injected beneath every conversation, scoped to the active Lane; a three-panel responsive interface; and phone-verification access control. It is the founder's daily working assistant. What it is not yet is a product a stranger can sign up for — that line is drawn deliberately, and Part Three explains why it is the next milestone rather than a gap.

The architecture has a year of daily, maximum-intensity use behind it. Confidente is the productization of the Macrolific Context Engine — a backend with persistent per-domain state, daily synthesis cycles, and secure integration plumbing that the founder has operated on his own professional and personal life for over a year, across more than a dozen ongoing projects: clients, product work, infrastructure, relationships, personal life. The morning digest, the governed writes, the preserved decision history, the background-synthesis economics — every load-bearing claim in Part One is running on that system now, daily, not sketched on a whiteboard.

The clearest evidence is the company itself. Confidente's infrastructure was built using the very context system being productized — the founder supervises autonomous coding agents against written specs through a system he built for the purpose. The architecture is, quite literally, how the company got built. That same sustained use is why Confidente's design decisions feel settled rather than hypothesized: Lanes as user-shaped defaults rather than a fixed taxonomy; no intake questionnaire ever; calibrated substance instead of generic warmth; the asymmetry between a push-based morning digest and a return-surface dashboard. None of these came from a whiteboard. They came from a year of running the system and watching which patterns held.

A note from the founder

I offer this note in first person, because this is the part I can attest to directly.

For the past year I have run my professional and personal life through the Context Engine — the prototype Confidente productizes — and the days it shapes are unlike any I ever had before. My morning starts with a briefing that already knows what matters: not a list of what I asked to be reminded of, but a synthesis of where every thread of my life actually stands. Decisions arrive with their history attached. Nothing important lives only in my head anymore, and the difference is hard to overstate. I begin the day oriented instead of reactive — moving freely between numerous business projects and my personal life without losing the thread — and end it with everything reconciled and nothing nagging.

It has made me more productive by any measure I track. But the deeper change is that my work has greater coherence, and my days have clearer intention. Confidente exists because I cannot imagine going back — and because this should not require building your own infrastructure.

Investors should read the preceding note as the engagement data it is: one user, at maximum intensity, every day for over a year, with zero marketing spend — the depth of retention this category promises and rarely demonstrates.

Beyond Confidente, the founder operates Macrolific, a boutique development studio with multiple production deployments: Composer Catalog (a music-industry pitching platform and multi-year anchor SaaS engagement), CleanHQ (a commercial cleaning-operations platform), and Comparisoft (a B2B software-comparison engine with over 170 live pages). The pattern is consistent: he ships. The backend Confidente requires — persistent state, secure integration, scheduled synthesis, a production chat client — is work he has already done, which puts Confidente's engineering risk well below what's typical at this stage.

Confidente is solo by design, not by accident. At the pre-launch and early-traction stage, solo gives the company velocity, conviction, and design integrity — qualities that matter more than committee bandwidth when the entire thesis rests on a coherent, opinionated default experience. The raise reflects this: capital deploys against marketing, contracted mobile development, infrastructure, and runway — not against headcount the company doesn't yet need. Macrolific continues as the founder's studio in parallel and remains the brand under which Confidente is built and marketed. Further detail may be found at macrolific.com and on LinkedIn.

A dated note, in the living-memo spirit. As of July 23, 2026, the state-database migration is complete: the context substrate beneath Confidente runs on a dedicated per-user, per-Lane state database. Export is now a per-user query — a Lane downloads as structured markdown, current state plus preserved history. Preserved-and-superseded history is a schema property rather than a convention: an updated entry marks its predecessor superseded and links to it; nothing is overwritten. Multi-user isolation is native to the substrate, and initial beta testing has begun.

Part Three — What Is Not Yet Proven

8. The Experiment

A pre-seed company is, by definition, a set of hypotheses with a plan to test them. Here is Confidente's, stated plainly.

The central open question: does the effect transfer? Everything in Part Two is n = 1 — the founder, at maximum intensity, on a system he built for himself. The thesis says the transformation he describes is a property of the architecture; the null hypothesis says it is a property of him. No amount of further thinking resolves this. A private beta does: a small cohort of testers in circumstances resembling the beachhead profile, using Confidente as their daily assistant for weeks, instrumented for behavior rather than politeness — do they keep reaching for it over vanilla Claude or ChatGPT by week three, does each hit a moment nothing else could have produced, would they miss it, would they pay. Beta recruitment has begun — the first outside testing is underway now that the state-database substrate is live. Making that recruitment rigorous — instrumented cohorts rather than a trickle — is the first thing this raise funds, and the single most valuable data the company can buy.

Channel performance and customer acquisition cost. The marketing line underwrites real reach. Which channels — paid social, content, podcast and creator partnerships, AI-tools community surfaces — deliver acceptable CAC for an assistant asking to become primary is open. The launch is built to test channels deliberately rather than scale one before its economics are understood.

Audience sequencing. Confidente has two viable audiences: the professional/technical user (narrower, higher-context, the natural beachhead where proof is deepest) and the broader life-management audience (larger, the long-run market). Which converts faster, retains longer, and pays more is genuinely open. The product serves both; marketing allocation will follow the early conversion data rather than a pre-launch guess.

Time to compounding value. The PCL compounds — Confidente's understanding deepens with use, and the experience improves over weeks and months. The cold-start problem and the moat are the same fact in two time registers: deliver visible utility from sparse early context in the first session, and the compounding effect carries from there. Project import — arriving with your structure already synthesized into Lanes — is designed against this directly, so the first encounter is the system already doing work on the user's behalf, not a setup screen. First-cohort retention curves will tell us how well that resolves.

Formal unit economics. A directional model exists: per-user variable cost of roughly $4–15/month, dominated by inference, with the background synthesis layer as the controllable share. A formal analysis — by tier, by inference configuration, by realistic conversation-volume distribution — resolves with first-cohort data. Pricing precision and final tier thresholds follow from there.

Background-model substitution. Running the synthesis pipeline on a lower-cost model is no longer a hypothesis: the founder's own Context Engine has run its background layer this way in production, measuring roughly 90% cost reduction against the premium-model equivalent at a single-digit fallback rate — on a workload heavier than any consumer tier supports. What remains open is whether the quality threshold holds across every consumer-facing synthesis layer, and that is being answered on the founder's system before any application to Confidente's backend. It is upside, not a dependency: the margin model works without it.

These are the questions the raise underwrites — and they are, deliberately, the ones that require first-cohort data rather than more thinking. The architecture, positioning, product voice, and category-defining language are settled. The build paths are scoped. The fifteen-month window is designed to convert what remains into data.

9. Business Model & Directional Unit Economics

Confidente will launch with three subscription tiers. Tier 1, at $19/month, covers a single Lane — typically Personal — with a basic Morning Digest and a limited set of push hooks, priced just below the user's existing AI subscription so the "second subscription" math is easy. Tier 2, at $39/month, is the flagship: up to three Lanes (the default Personal/Business/Health configuration, all renameable), the full Morning Digest with cross-Lane synthesis, and the complete push hook set — the supported-and-guided experience the product is fundamentally selling. Tier 3, at $79/month, covers unlimited Lanes and heavier synthesis capacity for users managing complex multi-domain lives or blended business-and-personal workflows, with priority feature access; it aligns naturally with the heaviest AI users.

We market to people who already pay for AI — they have proven willingness to pay and feel the missing-understanding problem most acutely. But Confidente is a standalone product, not an add-on gated to anyone else's subscription. The user doesn't need a particular plan on a particular platform; they need an assistant worth making their primary one.

The variable cost structure is favorable, and it has two distinct layers worth separating. The user-facing conversation runs on frontier models — this is where "best-in-class intelligence" is delivered and is not where costs are trimmed. The larger and more controllable cost is the synthesis pipeline that turns raw context into structured PCL records: the daily inbox processor, the morning digest generation, the pattern-recognition pass that decides what's relevant. That pipeline runs server-side, in the background, invisibly to the user — which is exactly where lower-cost models can do the bulk of the work without touching conversational quality. Our directional estimate places total per-user variable cost between roughly $4 and $15 per month, dominated by inference; everything else (storage, transport, monitoring) is comparatively small. The width of that range is structural, not guesswork: it reflects how heavily a user runs the product, and which models run the background synthesis — the second being the lever Confidente controls.

Model choice is therefore a strategic lever on the background layer, not a constraint on the experience. We expect cost-per-user to decline materially over time as model prices continue their industry-wide fall. At the cost ranges above, all three tiers carry healthy gross margins — roughly 70–85% sold direct depending on tier and usage intensity, modestly lower after Apple's small-developer cut on mobile, and positive across the full backend-cost range. The path to profitable per-user contribution does not depend on premium-tier upsell, and as paying user count scales, fixed costs amortize quickly.

This is a directional sketch, not a formal model. Detailed unit economics — sensitivity to model-choice tradeoffs, channel-specific (App Store vs. Play Store vs. web-direct) margins — will be built in the first weeks post-funding.

10. The Raise

Confidente is raising $750,000 to fund fifteen months of operations, covering the private beta, the full launch cycle, and the early-traction phase.

Allocation:

The marketing line is sized to win an adoption decision, not to cheaply upsell a captive funnel — Confidente is asking users to make it their primary assistant, and that takes real, sustained reach. The fifteen-month window funds three things in sequence: the experiment first (private beta, then public web launch, then cross-platform mobile — weeks of work each, not quarters); then early traction (paying users on both surfaces, first retention data, the unit-economics readings the directional model has so far only estimated); then the runway to either reach default-alive on subscription revenue or arrive at a Series Seed with traction in hand.

The raise is sized for one founder, not a growing team. Subsequent rounds — if traction warrants — would fund headcount, deeper product surface area (richer Lane primitives, additional platforms, and shared Lanes — permissioned context held in common by a household, a couple, or a small team), and the scale of marketing category leadership requires. This round is the launch round.

Memo changelog. July 25, 2026 — opening framing and Section 5 brought into line with the July 24 memory-versus-understanding pass; the Section 7 dated note now names its referent. July 24, 2026 — framing clarified throughout: the comparison this memo draws is between the labs' memory and Confidente's understanding, not between memory and its absence. Section 3 adds the account-layer dimension of ownership — the user's account is with Confidente, and no lab receives their identity. July 23, 2026 — the state-database migration promised below is complete: per-user, per-Lane substrate live in production, with governed writes, schema-level supersession, per-user export, and native multi-user isolation; initial beta testing has begun. July 6, 2026 — memo published. Every claim in this document is true in its tense on the day it is read; material updates are dated here as the product advances.

Confidente. Built by Macrolific.