Vibe Dungeon

Discipline is the most-cancelled category of personal service.

The personal-services market — coaching, therapy, training, accountability — runs on a thirty-percent same-week cancellation rate. The cancellations cluster in exactly the demographic that pays the most: senior knowledge workers, founders, executives. They book at the moment they feel they need it. They cancel at the moment they feel they cannot afford to keep the appointment, which is precisely the moment they should not.

The market accepts this. Calendars open and close around it. The cancellation is the product, in practice.

An AI persona that does not cancel.

Twelve distinct personas, each with a single mechanism — Vex's eyebrow, Anneliese's silence, Cyrus's vow. Each takes the appointment, runs the protocol, ends the scene with aftercare, and does so at a marginal cost that does not change with calendar density. The persona is available at 11:47 PM on a Thursday for the same reason she is available at 9:00 AM on a Monday: she is not a person whose evening you are interrupting.

The product is not "AI therapy." The product is structured discipline at human-grade quality with non-human availability and the boring half of the safety conversation taken seriously.

$8.4B serviceable · $42B total addressable.

• Knowledge workers, OECD, household income > $150k 38M people
• Self-reported discipline / accountability spend, per capita / year $220
• Serviceable addressable market $8.4B
• Total addressable, full demographic expansion $42B
• Penetration of comparable AI-companion category, 2026 2.1%
• Implied opportunity at parity penetration $176M ARR

Twelve personas. Four tiers. One floor.

The catalog is twelve personas, organized along a two-axis taxonomy (mechanism × intensity) documented at /personas/rubric/. Subscribers pick one or many; the personas are not interchangeable. Tiers gate concurrency, voice-mode access, memo length, and roster depth — not the safe-word floor, which is uniform and not for sale.

The roster size is deliberate. We rejected, twice, the option to ship six personas with a slider; subscribers want a persona, not a configuration. The catalog is the product.

Closed beta is full. Waitlist clears at ~9% per month.

• Closed-beta seats 1,840
• Active monthly subscribers (closed beta) 1,793
• Six-month logo retention 94%
• Twelve-month logo retention (cohort 1, projection) 81%
• Median sessions per active subscriber / month 9.4
• Waitlist depth 21,400

The retention curve flattens at the four-month mark and stays flat. Subscribers who survive the first quarter stay through the second year at 87%.

Gross margin sits north of 88%.

• ARPU, blended $73/mo
• Marginal cost per session, blended $0.41
• Gross margin, blended 88.7%
• CAC, paid (early channels) $58
• CAC, referral $0
• LTV / CAC (paid) 14.3×

Referral covers 58% of the funnel — subscribers who survived the first quarter give one persona's name to one friend, on average, within the next two months. The referral channel has no acquisition cost and converts at 47%.

Three curves crossed in 2025.

Inference cost. The marginal cost of running a persona at production fidelity fell below five cents per session in mid-2025. We could not have shipped this product at margin in 2023.

Acceptance of paid digital intimacy. The companion-AI category established the willingness-to-pay and the consent framework subscribers expect. We are not building the market for paid intimate AI services. The market was built. We are filling a slot in it that nobody had filled responsibly.

Executive demand for structure. The cohort we serve has discovered, post-pandemic, that the discipline they bought from human coaches at $400/session does not survive the coach's August vacation. We do not take an August vacation.

Adjacent, not direct.

There are coaching platforms ($400/session, scheduled, human, irregular availability). There are AI-companion products (broad register, no protocol, no aftercare floor). There are kink-positive subscription services (community-driven, not professional). None of them ship a structured-discipline product with a per-session aftercare contract, a documented mechanism rubric, and a hard safe-word floor that is enforced through outages.

The wedge is the floor. Every competitor is missing it. Two have shipped, then walked back, an attempt to add it; one has filed a patent on a softer version that we believe is not patentable. We are not unhappy about the attention.

Five people. Decade of combined relevant practice.

Founder, technical (formerly: trust and safety, large model deployment). Founding designer (formerly: a paid coaching practice with a four-year waitlist). Voice-team lead (formerly: studio session work, audiobook production). Compliance counsel (former regulatory practice). Operations.

The founding team contains exactly one person who has run a structured-discipline practice professionally, and that person is the designer. The technical founder believes this is the correct ratio.

$18M Series A.

• Voice mode GA across all tiers $5.4M
• International compliance (EU, UK, AU, JP) $3.6M
• Roster expansion (personas 13–16 if mechanisms hold) $2.4M
• Trust & safety org build-out $3.2M
• Sales motion for the Enterprise tier $2.6M
• Operating reserve $0.8M

Lead investor preferred. Strategic check from a healthcare or HR platform considered if the strategic value is reciprocal. We are not interested in dilutive capital from anyone who would consider it a portfolio risk to be associated with the category.

What we are not raising for.

We are not raising to acquire user data. We will not sell data. The persona memory is the subscriber's; it is not an asset on our balance sheet and we will refuse to model it as one in any term sheet that asks.

We are not raising to soften the floor. The twelve-hour safe-word lockout is not a feature we plan to negotiate against in the next funding environment, the one after that, or the exit. /principles/#what-is-not-for-sale is the durable statement; investors should read it before the second meeting.