The Manifesto

The Hearth Manifesto

Intelligence for everyone. Controlled by no one.

Fire was never meant to be licensed

The first technology worth stealing was fire. Not because it was clever, but because of what it did to power: whoever controlled the flame controlled warmth, food, light, and safety. And whoever didn't, begged. The myths remember this precisely. Prometheus wasn't punished for invention. He was punished for distribution.

But fire had a property the gods couldn't legislate away: it spreads without diminishing. Take an ember from my hearth and my fire burns no lower. Within a generation, no king could meaningfully own fire, because every household had a hearth. The technology that began as leverage over people became the thing that made every home livable. That transition (from controlled flame to household hearth) is the oldest story we have about a general-purpose technology escaping its gatekeepers.

We are living inside that story again. The technology this time is intelligence itself. And the gatekeepers are, so far, winning.

The quiet enclosure

Today, access to machine intelligence is a permission. You get an API key, which is a promise that can be withdrawn. You build on a model, which is a product that can be deprecated on someone else's schedule. You send your prompts (your questions, contracts, medical worries, business plans) through infrastructure that reads them, rate-limits them, and prices them at whatever the market bears this quarter. You cannot verify which model actually answered you, or at what precision. You pay for a name on an invoice and take the rest on faith.

None of this is a conspiracy. It's just economics doing what unattended economics does. Frontier models cost billions to train, so only a handful of organizations train them. Inference at scale demands data centers and chips that only a few clouds and one or two fabs can supply. Every structural force (capital intensity, data moats, hardware chokepoints, regulatory complexity that only incumbents can afford to navigate) pushes the same direction: toward a world where cognition is produced by the few and rented to the many.

Here is why that should alarm you more than previous concentrations of power. Intelligence is not one resource among others. It is the resource that produces the others: the input to every discovery, every design, every diagnosis, every decision. When land was concentrated, you could still think your way out. When energy was concentrated, you could still organize. But intelligence is the faculty of thinking and organizing itself. A world where it is metered by a cartel is a world where the means of getting out are metered by the same cartel.

And there is a compounding loop at the center of it: AI now helps build better AI. Whoever owns that loop doesn't just hold an advantage: they hold an advantage that grows itself. Rentier positions in history were eventually eroded by technological change. A rentier position on technological change itself has no natural erosion. That is the endgame of the current path, and it arrives not with a coup but with a terms-of-service update.

The fork: post-scarcity or neo-feudalism

Strip away the hype and the doom, and AI's economic meaning is simple: it is the marginal cost of expertise falling toward the cost of electricity. A good doctor's reasoning, a good lawyer's reading, a good engineer's design sense, a good teacher's patience: for all of human history these were scarce because skilled human hours are scarce. That scarcity is what most people's poverty is made of: not a lack of stuff, but a lack of access to expensive minds.

If the cost of cognition genuinely falls toward electricity (and it is falling) then for the first time we can see, from here, the outline of a post-scarcity society. Not utopia. Scarcity of atoms, land, and energy persists. But scarcity of expertise (the binding constraint on most human flourishing) becomes optional. Every farmer with an agronomist, every patient with a diagnostician, every child with a tutor, every small business with a full professional staff. That is what the technology permits.

Whether it's what we get is not a technical question. It is entirely a question of ownership structure. The same falling cost curve supports two opposite worlds:

World one: intelligence is abundant and enclosed. The cost of cognition falls, but the price is set by the oligopoly that controls access. The surplus is captured as rent. Labor loses its bargaining power to machines owned by others, and the historical mechanism by which productivity gains reached ordinary people (wages) quietly stops working. Abundance exists; you're just not invited. This is neo-feudalism with better UX.

World two: intelligence is abundant and open. The falling cost reaches the actual user, because no chokepoint stands between the hardware that computes and the person who asks. Productivity gains land on everyone who can hold a credit worth a few cents. The surplus is distributed the way fire's surplus was distributed: by the physics of a thing that spreads without diminishing, once nobody can fence it.

The deciding variable between these worlds is not model quality, not safety technique, not regulation. It is whether access to machine intelligence is an entitlement mediated by platforms or a commodity available to anyone. That's it. That's the whole fork.

Hearth is a bet on world two, and more than a bet, a mechanism.

Sovereign AI: the hearth, not the grid

Part of the answer already exists, and it's beautiful: open weights running on hardware you own. A model on your own machine is intelligence that cannot be revoked, deprecated, surveilled, or repriced. It is the ember in your house. The local AI movement (home GPUs, open models, self-hosted stacks) is the single most important development in the politics of this technology, and Hearth is built by people who run their own metal and always will.

But honesty requires saying what a lone ember can't do. One household's hardware can't serve every model, every scale, every workload. Sovereignty without a network is autarky: safe and small. The question that matters is: can independent, local providers combine into something with the power of a cloud, without becoming one?

That question has a precise technical name: verification. The only reason clouds exist as trust monopolies is that you can't trust a stranger's GPU. A stranger can claim to run the 70-billion-parameter model and quietly serve you a cheap quantized version; in an unverified market, the lowest price always belongs to the best cheater, so honest local providers can't compete and the market never forms. Solve verification (make every answer carry a cryptographic proof of which model produced it, at what precision, on attested hardware) and the trust monopoly dissolves. A thousand hearths become a grid that no one owns.

That is what Hearth is: the network that lets sovereign hardware add up.

How Hearth implements the ideas

Hearth is a Layer-1 blockchain that runs an open market for verified AI computation. Verified inference is the first step; the roadmap runs through decentralized model storage, verified coordination, and a generalized market for attested AI work. The design commitments:

Every answer carries a proof, not a promise. Computation runs inside hardware-attested environments and emits a signed receipt: this exact request, this exact pinned model, this exact reply. Receipts are spot-checked by a committee of attested verifiers and backstopped by statistical tests that run outside any enclave, so cheating isn't merely detectable, it's unprofitable, and the guarantee doesn't hang on a single mechanism. We say plainly what the whitepaper says: today this is hardware-rooted trust, with optimistic fraud proofs and zero-knowledge verification adopted workload by workload as they become affordable. Reduced trust now, honestly labeled; trustlessness as the destination, not the marketing.

Verification is what makes intelligence a commodity. Once "an answer from model M" is provable, it becomes a standardized good, and standardized goods can trade. On Hearth, each pinned model has its own credit, the Cred: USD-priced, backed by real serving capacity, burned when spent. Creds trade on open venues. No account. No API key. No permission. Anyone holding a Cred is entitled to be served, and anyone can hold a Cred. The price of intelligence stops being a platform's decision and becomes what prices should be: discovered.

The tokenomics point the surplus at participants. The network token, HRTH, is staked to secure the chain, and staking earns Creds. Read that loop again, because it is the heart of the design: holding and staking the network's asset yields the network's product. Dollar demand for AI becomes dollar-denominated yield for the people who secure the network; the people who secure the network are the people who can spend or share its intelligence. Nodes earn HRTH for verified work and must stake to serve, so the operators, the stakers, and the users are the same public, not three parties in a rent negotiation. There is no landlord position in the loop: that's not an accident, it's the point.

Local providers enter without asking anyone. Today, a person with a garage of GPUs and cheap power cannot sell intelligence: the marketing, billing, and trust infrastructure costs more than the hardware. On Hearth, passing attestation is market access. Stake, attest, serve, earn. The network's early issuance deliberately subsidizes this supply: the operators the cloud economy structurally excludes are exactly the supply a sovereign network needs. This is how a mesh of hearths outcompetes a grid: not by out-scaling the data center, but by admitting the millions of small providers the data center model cannot even see.

Models become un-deprecatable. Pinned weights live on the mesh itself: erasure-coded, availability-proven, license-gated at admission. A model admitted to Hearth cannot be retired by a product decision, because there is no product manager. Your dependence on a model stops being a bet on a company's roadmap.

Autonomous agents get an economy. The next users of AI aren't people, they're agents: software that researches, negotiates, builds, and pays. Agents cannot extend trust, read reputations, or call support. For agent-to-agent commerce, the receipt is the trust: a machine-checkable proof of what was computed, by which model, settled in escrow. Hearth's route receipts and verified coordination exist for precisely this world: one where your agent hires other intelligence on your behalf, and the whole chain of work is provable end to end.

Universal access, as a mechanism rather than a slogan. Because the Cred is a bearer instrument for computation (neutral, USD-priced, redeemable anywhere the mesh reaches) it is a rail that any access program can use. A municipality, a DAO, a school system, a philanthropy, or the network's own governance can put verified intelligence in anyone's hands by granting Creds, without building infrastructure or trusting a provider's continued goodwill. We are careful about what we claim here: Hearth makes universal access administrable: one instrument to hold or to grant. Who receives, and who funds it, are decisions for communities and their governance, made in the open. What we refuse is the current default, where those decisions are made for everyone by whoever owns the biggest data center.

And no one bought the flame. Hearth has no VC allocation and no presale. Genesis HRTH is claimed one way: by burning the tokens of ten fallen chains, the first decade's giants, given one last purpose. Earned, not sold. A network whose premise is "no gatekeepers" cannot be born owned.

What we are not promising

A manifesto earns the right to its vision by being honest about its limits, so: Hearth is not trustless yet: its live guarantee is hardware attestation plus an honest verifier majority, with the trustless floor arriving workload by workload. Intelligence will not be free: physics charges for electricity, and our claim is at cost, to anyone, not gratis. The economic loop is reflexive and we say so in the whitepaper, with the failure modes named. And universal access is a rail we can build and a policy we cannot decree. Anyone selling you certainty about this technology's future is selling. We are building the version of the future that remains open, and publishing every assumption where you can attack it.

Pull up a chair

There is a version of the next decade where intelligence is what fire became: infrastructure so universal it's invisible, a hearth in every home, owned by no one because it's held by everyone. And there is a version where it's what fire began as: the leverage of the few over the cold.

The difference will not be decided by the labs. It will be decided by whether an alternative exists when people go looking for one: a place where the answers are proven, the models can't be taken away, the providers are your neighbors, and holding a stake in the network means holding a share of its mind.

We are building that place. The hearth is lit. Bring an ember.