Spiritual Sovereignty

The vision

We are building the meditation infrastructure for AI to scale awakening.

The Sacred Act observes a person, understands their state through what they say, how they say it, their biometrics and their environment, prescribes the technique that fits that state, and guides them through it. It learns the person's mind and body over months, so the guidance in the sixth month is shaped by the five behind it.

The mission is spiritual sovereignty. Said plainly, in a demo to four people who have known me my whole life:

There should be access to learn about it and to practice it, in various ways, from who you are and from how you need it in any moment. And for this to learn you over time, so that you aren't stuck in only one way of thinking ever. The choice to open up to more, or to close to one tradition, is yours.

Sovereignty is not access to a library. A person alone with a catalogue of recordings is not sovereign over their practice, they are browsing. Sovereignty is having something that knows what you need and can hand it to you, on your terms, in the moment you need it. That is what a teacher does, and it is the one thing the category has never sold.

And it does not end at the phone. The practice is meant to be reachable where a person actually is — in a car, on a plane, in a room with other people, or through whatever assistant they already talk to. The app is how the layer gets built and proven. It is not the ceiling of the thing.

Where we are right now

Nine months of building, an app in testers' hands, and three technologies that did not exist before we made them.

The app is in TestFlight, not on the App Store. It generates real meditations for real practitioners today. What testers have said about it, verbatim:

The start of my day has been so clear and productive. I feel like a light worker. I don't have this weight of inertia or procrastination.

The guided meditation: the words and how they are put together is so powerful. I thought to myself: I don't even know some of these words, but my body does.

This application is otherworldly and made my sister cry.

Three things had to be built that did not exist. They are the company's actual assets.

What is still open. The app is not generally available. 650 of the 1,419 techniques do not yet have their finished artwork. The instrumentation that tells us whether a practice landed is incomplete, which is what keeps the learning loop thin — and fixing it is the highest-leverage work outside the launch itself.

The roadmap

Four phases, each gated on evidence from the one before it.

The sequencing rule is that nothing expensive happens before the thing that would justify it has been measured. Dates are targets set against the current open work, not commitments.

Phase 1 — Soft launch to 100+ practitioners

The organizing goal. Everything in it blocks launch; nothing outside it does.

Three workstreams run in parallel: the subscription wall has to hold before strangers arrive, the service needs crash reporting, error alerting and a durable cost meter so a bad day announces itself, and the App Store submission needs its portal work — both subscription products created, privacy labels completed, a review account provisioned, the version scheme corrected.

Pricing is decided and locked: $199 a year with a seven-day free trial, or $19 a month. The go-to-market section explains why that number and not the category's $69.99.

The output of this phase is not revenue. It is evidence: which communities convert, what the real retention curve looks like, and whether a person who finishes a first session comes back for a second.

Phase 2 — Public launch, margin, and the voice

The Sacred Act is a generative app. Every session costs real money to produce, where Calm and Headspace serve a catalogue at effectively zero marginal cost. That is a structural difference and it gets engineered against, not hoped away.

Four fixes are identified. Together they cut cost per session by more than half.

1. Pre-generated technique art. ✅ Landed. The app no longer generates images at all — it looks up a rendered relic per technique. This was the single largest variable cost, roughly 40% of the unit cost, and it is now a fixed library rather than a per-session spend. What remains is content: 769 of 1,419 techniques have a complete light-and-dark pair, 650 do not. That is a deliberate deferral of spend, rendered in waves as it is paid for.
2. Gate the practitioner-portrait rebuild. Three code paths currently rebuild a person's deep portrait on activity that does not warrant it, at roughly 33 seconds and a large model call each time. Gating it is a 95%+ reduction on that line.
3. Context-cache the technique JSON. About 2,000 tokens of authored technique data go into every generation. Cached context is priced at a tenth of uncached input. The JSON itself is never compacted or trimmed — it is the accuracy of the thing — so we optimize around it rather than through it.
4. Move off preview model IDs. Ten call sites still point at preview builds where a generally-available equivalent exists.

Our own voice model

Voice is the largest remaining vendor line on every session, and it is the next thing we own.

Today the guide is synthesized by an outside vendor, billed per character, with a hard ceiling on how many voices can be generated at once — we run a circuit breaker in production specifically because of that ceiling. Every practitioner who starts a sit at the same moment as another is competing for the same rate limit.

Training our own voice model changes four things at once:

• Margin. A per-character vendor cost becomes fixed infrastructure. This is the single largest step toward per-session profitability after the art.
• Concurrency. The ceiling disappears. Ten thousand people can begin a meditation in the same minute, which is the difference between a product and a live group event at scale.
• Speed. Generation latency drops, and latency is felt — a person waiting to be guided is a person leaving.
• A later revenue stream. A voice model trained specifically for contemplative speech — real breath, structural silence, pacing that follows a practice rather than a script — is an asset other people will want to license. No general-purpose TTS is built for this, which is exactly why we had to measure its limits ourselves.

Phase 3 — The learning loop and the scholars

This is the phase that makes the product compound.

The prescription can only learn from sessions where we know which technique was practiced and whether the person finished. Completion marking did not exist for the first six months of the app's life. Fixing the mark fixes four things at once: session counting, the biometric backfill, the learning loop, and course progression — because unlocking the next class is derived from completion. The ranker's weights get recalibrated from real outcomes once months of that data exist, not before late 2026.

The other half of quality is human. We are hiring PhD scholars in the contemplative traditions to review the corpus line by line — flagging misattributions, syncretic blendings, oversimplifications, and any place a technique has been stripped from the ethical context that made it work. A machine can structure 1,419 techniques. It cannot tell you when sati has been flattened into "mindfulness" and lost the discernment that made it sati. A scholar who has published in the tradition can.

That review is a product improvement, a credibility asset, and — as the next section explains — a content engine.

Phase 4 — The practice layer, everywhere

Two things live here.

Group meditation. Several people, each on their own device, in one shared space, each adding their own input, and the system generates a single guided meditation for the group. This is where the voice model's concurrency stops being an economics story and becomes a product that could not otherwise exist.

Every surface a person already sits still on. A meditation is one of the few things people genuinely want during the exact moments technology currently offers them nothing: the back of an autonomous vehicle, a commute, a long flight. We intend to be in CarPlay and in the native systems of Tesla and Mercedes, in Waymo, and on the seatback screens of commercial aircraft and private jets. These are not app-store downloads competing for attention — they are captive, unoccupied, often stressful stretches of time where a guided practice is the most useful thing the screen could offer.

Alongside them sits the infrastructural claim: the corpus reachable by any assistant a person already talks to.

How we fund it

Founder capital to build it, a small round to prove it, institutional capital only once the evidence exists.

This section describes our capital strategy. It is not an offer to sell securities and nothing on this page is an invitation to invest.

Everything above was built without institutional capital. Nine months, 2,661 commits, three novel systems and a working app, funded by the founder. That is the base case we argue from: the expensive part is already spent.

The funding logic follows the phases, because a generative app fails differently than a catalogue app. A catalogue app's risk is whether anyone wants it. Ours is that and whether the sessions can be served profitably at scale. Capital buys the answer to the second question, not just runway.

What the soft launch actually costs. The non-inference spend to get from TestFlight to a full public launch is roughly $25,000, and it is unusually legible:

The largest line is the one without a fixed price: every meditation the first cohort generates is paid for by us. That is the point of the phase — the sessions are how we learn which communities value this, and they are not free to produce.

The discipline. Paid acquisition does not run before the retention curve is known, because nineteen in twenty people who install a meditation app stop opening it within a month, and buying installs into that curve is how the category burns money. Institutional capital gets raised against measured conversion and measured cost per session, not against a deck.

The markets, and why

Six markets. The largest one does not exist yet, and that is the one we are building for.

The macro picture is favourable and largely beside the point.

Sources: Global Wellness Institute 2025 Monitor, Grand View Research, Precedence Research, International Coaching Federation 2025 Global Study. The spread on the meditation-app growth rate is the most honest fact in this table: the category is poorly measured, and any single CAGR quoted as fact is a vendor's model.

We are not trying to take share in a $2.5 billion category. Six markets matter to us, in this order.

1. The fifty-one million who intend and have no practice

This is the market, and nobody serves it. 51 million American adults hold an intention to meditate and have no practice — 19.1% of US adults. About 30 million have one. The people who intend outnumber the people practicing by better than one and a half to one, and every meditation app on earth competes for the smaller group.

Awareness is not the barrier: 92% of Americans know what meditation is, and roughly half have tried it at least once. Information is not the barrier either. Guidance is the barrier, and guidance has been rationed by the number of teachers who exist — the largest certification program in the field has produced about 8,000 of them, across more than seventy countries, for those 51 million people.

Converting 1%, 3% or 5% of the intenders at $199 a year is $101M, $304M or $507M in annual spending that does not exist today. None of it is taken from Calm. We walk through that model, and its largest untested assumption, in Creating the Meditation Market.

The precedent is not speculative. When language instruction became software, four in five of the people who showed up had never been in that market at all. When software engineering got an AI assistant, 36 million new developers joined GitHub in a single year, 80% of them using an assistant in their first week. Both categories were gated by a scarce expert. So is this one. The disanalogy is real and we name it: code has an objective pass condition and a meditation does not, so the feedback loop we need has to be manufactured out of completion, self-report and physiology. That is phase three.

2. The premium buyer who has never been offered software

The category's price ceiling is about $5 to $6 a month, measured on people buying a sleep aid. The same population pays $420 to $980 for four sessions of personal instruction. Those are two different buyers, and only one of them has ever been sold to by an app.

People pay ten times more for a person than for a library, consistently, and the supply of people cannot scale. That is the arbitrage.

3. Coaches and retreat leaders — the B2B product

123,000 practising coaches have no way to give their students a real practice between sessions. This is the clearest unmet need we have found, and it is a business on its own.

A coach running a virtual meditation retreat today has two options, and both are bad. They send YouTube links — generic, unpersonalized, ad-interrupted, and a direct handoff of their student's attention to someone else's platform. Or they record their own meditations, which does not scale past a handful, cannot adapt to the individual, and turns a coach into an audio producer.

Neither option tells the coach anything. They cannot see who practiced, who stopped, who is struggling, or whether the arc they designed is landing. They are teaching blind between sessions, which is exactly where the practice either takes or does not.

We already have the pieces. The app builds Paths — a scheduled arc of practices planned backward from an intention, read in named stretches with a milestone at the end of each. Sold to a coach, that becomes a custom Path for their cohort, generated fresh for each student against their own state, with a dashboard showing the coach who is practicing and how far along they are.

The coach keeps the relationship. We supply the practice and the instrumentation. It is a seat-based product with a built-in distribution motion, because every coach arrives carrying a cohort.

4. Lineage communities

A person practicing inside a living tradition is currently served worst of all, because a general catalogue has to flatten every tradition into a house style to be sellable. Our corpus does the opposite — techniques stay attached to their lineage, their source text and their contraindications.

That makes depth-by-lineage a real go-to-market unit. 64 Islamic practices from 19 works, with the Qur'an readable in the app, is not a content category, it is a reason for one community to trust the product on its own terms. The same construction is available for every tradition in the 249, and the scholar review in phase three is what makes that trust earned rather than claimed.

5. Enterprise mental health

Employers are the largest buyer of mental-health software, and they buy on evidence, not on feel. Getting there requires something the category mostly skips: real studies.

We are hiring a PhD researcher in contemplative neuroscience to design and run a study end-to-end on what these generated meditations actually do — measured, and published whichever way it comes out. That work has three payoffs. It tells us whether the prescription is doing what we believe it does. It is the evidence base an enterprise procurement process asks for. And it is the path to being approved as a recognized mental-health benefit rather than a wellness perk, which is the difference between a line item and a rounding error.

This is the longest-dated market on the list and the one that most requires patience. It is also the one where a structured, sourced, contraindication-aware corpus is a regulatory asset rather than a nicety.

6. The practice layer, and every surface

Apple reported more than 2.5 billion active devices in January 2026 and rebuilt its assistant layer on frontier models in June. Whatever that layer can do becomes what a couple of billion people can do without deciding to go looking.

Those assistants will be asked for spiritual guidance. They currently answer from a blur of secondary sources with no lineage, no contraindications and no idea who is asking. A structured, sourced, safety-annotated corpus of technique is the thing that is missing, and it is exactly what we have been building.

The same argument extends to every screen a person sits still in front of. Automotive and in-flight entertainment are distribution channels nobody in this category has taken seriously, and they hold precisely the audience the category claims to want: people with time, no task, and often some stress. This is the largest market and the least certain.

How we go to market

Evidence first, then rooms and creators, then paid — in that order, on purpose.

Prove it small

The soft launch to 100+ practitioners answers which communities convert, at what cost per session, with what retention. Every later channel is priced off those answers.

Research and scholarship as the top of the funnel

We publish what we measure, including the results that do not flatter us — the model benchmark that found our own prompt was hurting output, the voice study with a negative result in it, the market report showing willingness to pay in this category is low. This filters for the practitioner we want: the one who reads before they subscribe.

The scholar program extends this into something better. As each tradition's corpus gets reviewed, we record the scholars talking through it — what the technique is, where it comes from, what the text actually says, what modern practice gets wrong about it. That is a podcast nobody else can make, because nobody else has both the corpus and the scholars. One body of work that improves the product, proves the depth, and produces months of creative at the same time.

Live meditation in real rooms

Group meditation events, in partnership with wellness studios and health resorts. Three things happen at once: new practitioners are onboarded in the moment they are most open to it, the brand gets positioned beside places people already associate with serious practice, and we come away with the creative — a room of people meditating together is the footage, and it cannot be faked.

The format is proven. Fashion and wellness brands run exactly these events with editorial partners, and they work because the experience is the marketing. Contemplative practice spreads through rooms and through people, not through impressions. A private launch event for our first practitioners in San Francisco is the first of them.

Creator distribution

Launch creative is made by an agency rather than by the founder, and distributed through partners with an aggregate reach on the order of 100 million. Buying it rather than making it returns roughly six weeks of founder time to the product during the window that needs it most.

Coach partnerships

The B2B product is also an acquisition channel. Every coach who runs a cohort on our Paths brings their students into the app under their own relationship, at a customer-acquisition cost of effectively zero. We would rather earn a hundred coaches than a hundred thousand impressions.

Paid, last

App Store and Meta, after the retention curve is known. Not before.

Price as positioning

$199 a year against a category rate of $69.99 is a claim, and it has to be earned inside the first session. Across thirteen premium products we measured, price is absent from the landing page in eight, and the two highest-priced meditation products hide it hardest — the more expensive the product, the later the price appears. $199 is still roughly a fifth of what a person pays once for four sessions of personal instruction.

What we do not claim

The honest limits of everything above.

That the market-creation model is a forecast. It is a model. Its largest assumption — that removing the guidance constraint converts intenders into practitioners — has never been tested by anyone, including us. Testing it is what phase one is for.

That the analogies are evidence. Language learning and software were remade by removing a scarce expert. That is a reason to look, not a proof.

That personalization currently beats a teacher. It does not. The prescription's weights are hand-set and the data that would tune them does not exist yet.

That the roadmap dates are commitments. They are targets against known open work. The sequencing is the part we would defend, not the calendar.

That we have retention. We have testers, we have sessions, and we have people who were moved. We do not yet have a retention curve, and anyone who tells you they know how this performs at 100 users — us included — is guessing.

The one sentence

The Sacred Act exists to scale awakened awareness by making the practices of humanity's contemplative lineages accessible and personal to the modern mind. This is the responsibility of stewardship in the age of technology.