The HALO whitepaper is here
The full design of the open inference economy — SPEX verification, the fee-to-buyback loop, staking, emissions, genesis, governance. And no venture allocation.
Intelligence is becoming the most important resource of this century. Right now, access to it runs through a handful of companies.
To use a frontier model today you need an account that can be closed, an API key that can be revoked, a card issued in an approved country, and continuous compliance with terms that change without notice. Whole regions are fully blocked. Agents, the fastest-growing consumers of inference on earth, can’t open accounts without a human willing to sponsor them.
And the control is tightening, not loosening. Governments are moving to ban access to specific models. Providers push their best capabilities behind enterprise tiers, waitlists, and jurisdictions. It is no longer a stretch to imagine a world where most people are locked out of the best intelligence available: by law, by geography, or by a policy written in a boardroom they will never see.
Halo exists for exactly that world.
What Bitcoin did for money, Halo does for intelligence
Bitcoin took a resource controlled by institutions and made access to it a property of a network that nobody owns. No permission to join. No authority to freeze an account. No border at which it stops.
Halo applies that same construction to inference.
Every operator on the network is a doorway to the models they can reach. One resells frontier-API capacity. Another serves an open-weights model from a machine in their spare room. A hundred of them together become a single marketplace for any model. And the crucial property is this: if any one operator can reach a model, everyone on the network can.
Supply is permissionless, so anyone can register and start serving. Demand is permissionless, so anyone holding USDC can buy. No account, no KYC, no geography. A student in a sanctioned country, a founder deplatformed overnight, an autonomous agent with a wallet and no legal identity: to Halo they are all the same thing. A paying consumer.
That is censorship-resistant intelligence. Access secured not by the goodwill of a platform, but by the structure of the network itself.
Halo is already live on Base mainnet today: real operators, real USDC, settled onchain.
What we have learned
An open network with pseudonymous participants has to answer two hard questions, and the whitepaper is our full answer to both.
How do you know the operator actually ran the model?
Not with trusted hardware. Not with zero-knowledge proofs that would cost more than the inference. With SPEX, statistical proof of execution. Model outputs are fingerprinted as Bloom filters, roughly a kilobyte of compressed token IDs. An independent verifier re-runs the inference and tests its own tokens against that fingerprint. An honest run overlaps almost completely. A fabricated one lands at the filter’s own false-positive floor, indistinguishable from noise. There is no ambiguous middle, and a check takes seconds.
Then it scales. A verification task fragments across hundreds of micro-verifiers, each testing a slice, hundreds of independent checks completing in under three minutes. Trust in the network isn’t delegated to an authority. It’s checked, continuously, by the crowd.
How does a network with no owner pay for its own integrity?
That’s the job of HALO, and it’s worth being precise about what the token does and doesn’t do. Nobody who merely uses Halo ever needs to touch it. Consumers pay USDC. Operators earn USDC. Stable pricing is what makes an inference market usable, and we’re not sacrificing that to make a token look busy.
HALO lives one layer above, with the actors who keep the marketplace honest:
- Trust roles stake it. Verified operators, SPEX verifiers, and eventually federated relayers each post HALO to hold their role. Everyone whose honesty keeps the network open has capital committed to keeping it open.
- Revenue buys it. A protocol fee is collected on every settled job. The bulk of it feeds a programmatic onchain accumulation; the rest funds a USDC treasury that pays for operations without ever selling tokens. Most of the HALO bought will be available as incentives for network participants, and the rest is burned, on a governance dial with published bounds.
- Work mints it. New supply is emitted only against verified usage, inside a fixed budget that shrinks every year on a published schedule. The cap is a ceiling, never a promise: each epoch mints the lesser of the schedule and a multiple of volume actually settled. If the network doesn’t work, the network doesn’t print.
There is no venture allocation. Not a reduced one. None.
And instead of claiming to be deflationary, the paper publishes the exact settled volume at which it becomes so, and commits to a live dashboard counting toward it. Those numbers are falsifiable on purpose. We’d rather be measured than believed.
The best crypto UX is no crypto UX
There is a version of this network that only ever serves people who already own a wallet, already hold stablecoins, and already enjoy reading about settlement design. In the grand scheme of things, that version is a rounding error.
What we’re building for is millions of people who will never learn what a Bloom filter is, never open a block explorer, and never find out their prompt was paid for onchain. The blockchain here is infrastructure. Nobody checks which undersea cable carried their video call.
So the crypto is engineered to disappear:
- An account that meets you where you are. It is an onchain account, which is what makes it genuinely yours and impossible for anyone to close. How you open it is your choice: connect a wallet if you have one, or use a social login or a passphrase if you don’t. Same account, same guarantees, whichever door you walked through.
- Fund it with whatever you already hold. Pay in with a card. Or send crypto from whatever chain it happens to be sitting on, and it arrives as USDC on Base automatically. No bridging, no swapping, no working out which network you’re on or why that matters.
- No gas. The facilitator sponsors transactions on the user’s behalf. Consumers never hold ETH, never top up for fees, never watch a request fail because a gas estimate moved.
- No signing per prompt. You fund a vault once and register a session key. After that, work is covered by off-chain receipts that cost nothing to produce. No wallet popup standing between a question and its answer.
None of this is a different product for different people. It is one network with as many front doors as it takes: the crypto-native user gets the appropriate one, everyone else gets the one that looks like software they’ve used before, and both end up in the same market with the same protections.
And this is the part that matters: abstracted is not custodial. Hiding the machinery is trivial if you take people’s keys and run a database behind a friendly logo. That is just the old arrangement wearing new words, and it fails the moment someone decides you shouldn’t have access. The harder build, and the only one worth doing, is a product where a user never encounters the concept of a private key and still holds their own funds, still cannot be deplatformed, still owns an identity they never had to apply for.
That is the distribution thesis in one line. Sovereignty is what makes Halo worth using. Simplicity is what makes it usable by everyone. Crypto-native users get a permissionless market they can inspect down to the contract; everyone else just gets AI that works, from anywhere, without an account, at a price set by competition rather than by a committee. They never need to know why it’s possible.
What we’re actually trying to build
A marketplace where the price of intelligence is set by a global pool of operators competing on cost and reliability, instead of by the pricing committee of whichever company happens to own the model.
A network where an agent is a first-class economic citizen, able to buy the compute it needs with a wallet rather than a corporate sponsor.
A rail where the machine that checks the work is the crowd, and the machine that funds the checking is the traffic itself. Usage pays for honesty; honesty attracts usage. That loop, running without anyone’s permission, is the whole thesis.
The whitepaper walks through all of it: protocol architecture and settlement design, SPEX and swarm verification, the fee-to-buyback loop, staking roles and timing, the full genesis allocation and release schedule, emissions, and the narrow, slow governance that can change any of it.
We’ve also tried to be honest about what isn’t finished. One routing layer is still operated by us. The roadmap federates it into a stake-gated set of independent relayers, and until that ships it remains the network’s last chokepoint. The first staked roles arrive in H2 2026, with verifier and relayer staking tied to their own rollouts.
Read it, then use it
Intelligence that belongs to no one has to be built by a lot of people. Here’s where to start:
- Read the whitepaper. The complete design, top to bottom.
- Use Halo. Fund a vault, pay per prompt in USDC, no account required.
- Run an operator. One command, serve up any model you can reach, and get paid.
- Read the source. All of it is public.
The network is live. The paper explains where it’s going. Halo is open to anyone who wants in.
Nothing in the whitepaper or this post is an offer, a solicitation, or investment advice. HALO confers no claim on revenue or assets except as implemented in published smart contracts. Digital assets involve substantial risk of loss.