n1o Whitepaper
A complete AI assistant backed by a worldwide network of consumer GPUs. You pay for the inference you actually use; the people who own the hardware get paid for the work they actually do.
01 / ProblemAI runs into a wall of compute
Modern AI is gated by one thing above all others: the cost of running the models. Training gets the headlines, but the recurring bill is inference — every answer a model generates consumes GPU time, and GPU time is scarce and expensive. High-end accelerators are concentrated in a handful of clouds, allocated first to the largest buyers, and priced accordingly. For most builders and users, capable AI is either rate-limited, costly, or both.
At the same time, an enormous amount of capable hardware sits doing nothing. Millions of consumer GPUs are idle for most of the day — in gaming PCs, workstations, and, notably, in the mining rigs that were left stranded after Ethereum moved off proof-of-work. When Ethereum switched to proof-of-stake in 2022, the single largest use for consumer GPU hashpower disappeared almost overnight, leaving warehouses and bedrooms full of cards with no economic job to do.
So we have two opposite problems at once: AI is short on affordable compute, while a global fleet of perfectly good GPUs is short on useful work. n1o exists to connect the two.
02 / SolutionOne assistant, many GPUs
n1o is a complete AI assistant — available on the web and as a desktop app — that runs on a distributed network of consumer GPUs executing open models. To a user it feels like any modern AI product: you ask, it answers. Underneath, the inference is served by real machines contributed by real people around the world, rather than by a single centralized data center.
Using open models is deliberate. It keeps the stack inspectable, avoids lock-in to any one vendor, and makes it possible for the compute layer to be operated by a community instead of a landlord. The assistant is the product people use every day; the network is the engine that makes it affordable and resilient.
n1o is designed to become open-source and to decentralize progressively, in phases. We start centralized so the product works reliably from day one, then hand responsibility to the community step by step — until people run the nodes that power the network and earn for the compute they provide. The goal is not a company that rents you AI, but a network the community operates and shares in.
03 / The minerGPU-on-demand, paid for work
The supply side of n1o is the miner. You install the miner, connect a GPU, and it joins the network as an inference worker. When the network routes a request to your machine, your GPU runs the model, returns the result, and you get paid for that unit of completed work.
This is the important distinction: you are paid for inference performed, not for merely being online. We call it Proof of Inference — rewards follow verifiable, useful computation delivered to a real user, not idle connection time or wasted hashing. A GPU that does more real work earns more; a GPU that sits connected but does nothing earns nothing. Compute is spent on answers people actually want.
n1o is a second life for post-merge mining hardware. The cards that lost their job when Ethereum left proof-of-work can be put back to productive use serving AI inference — and the settlement, accounting, and payouts for that work live on Solana, chosen for its low fees and fast finality. In short: revive stranded GPUs, and migrate that hashpower niche onto a chain built for high-volume micro-settlement.
04 / CreditsHow value moves through the system
The economic loop is intentionally simple:
- Users buy credits to use the assistant. Credits are priced in a stable unit of account — USDC — so a credit's value is predictable and not tied to a volatile asset.
- As you use the AI, credits are consumed in proportion to the compute your requests require.
- Miners get paid for the inference they serve.
- The platform keeps a small spread to fund routing, settlement, safety, and development.
That is the whole machine: buy credits, use AI, hardware owners get paid, the platform takes a thin margin. It works today with USDC and needs nothing more to function.
A native $n1o token is a possibility being explored for the future. It is not launched, not promised, and may never exist. There is no price, no sale, and no expectation of profit attached to it. Nothing here is an offer of a security or an invitation to invest. Credits work with USDC today, regardless of whether any token ever ships. If a token is ever introduced, it would be a utility within the network and only if it is genuinely justified — see the roadmap and risks below.
05 / The spreadOne transparent margin, not a double fee
There is a single platform spread of roughly 15–20% between what a user pays for a unit of AI and what the miner receives for producing it. That is the platform's entire take. Users are not charged a fee and then miners charged another fee on top — it is one margin, applied once, sitting between the two sides of the same transaction.
Why one spread instead of layered fees? Because it is honest and easy to reason about. A user can look at what they pay, a miner can look at what they earn, and the difference is the platform's share — nothing hidden in between. That spread pays for the parts of the system nobody sees but everyone relies on: request routing, result verification, settlement on-chain, abuse prevention, and continued development of the assistant and the network.
06 / Miner economicsHonest about earnings
We will not pretend miner income is fixed or guaranteed, because it is not. Earnings vary — they depend on how much real work the network sends your way, your GPU's capability, uptime, energy cost, and overall network conditions such as how much user demand exists relative to how much compute is online.
What we can state plainly:
- You are paid for completed inference, priced by the work done.
- More capable, more available hardware tends to receive more work — but nothing is promised.
- When demand is low relative to supply, there is less work to go around, and earnings fall accordingly.
- There is no guaranteed return, and you should treat running a miner as providing a service, not as an investment.
Run the numbers against your own electricity price before committing hardware. n1o gives idle GPUs a chance to earn; it does not promise that they will.
07 / Network layersRoles and payout split
The network is made of a few cooperating roles. People can choose to run one or more of them and earn for the part they contribute.
- GPU workers — the machines that actually run the models and produce inference. This is where the compute happens and where most of the reward goes.
- Relays — nodes that move requests and results across the network, improving reach, latency, and resilience.
- Pool coordinators — nodes that organize workers into pools, distribute jobs, and aggregate results, taking a small coordination fee.
- Protocol / platform — the layer that verifies work, settles payments, and maintains the assistant and the network software.
| Role | Indicative share | What it does |
|---|---|---|
| GPU workers | 65–75% | Run the models and produce the inference users pay for. |
| Protocol / platform | 15–20% | Verification, settlement, safety, and ongoing development (the spread). |
| Relays | 5–10% | Route requests and results; add reach and resilience. |
| Pool coordinator fee | 1–3% | Organize workers, distribute jobs, aggregate results. |
Ranges are indicative and overlap by design — they describe intent, not a fixed contract. The principle is durable: the majority of value goes to the people running the GPUs.
08 / RoadmapPhases, not dates
The following are intentions and a direction of travel, not commitments to specific timelines. We ship in phases and only advance when the prior phase is solid.
Phase 1 — A product that works
A centralized platform with the assistant, credits paid in USDC, and miner onboarding. The priority is reliability: real users getting good answers, and real miners getting paid for the work they do.
Phase 2 — Spread the compute
Distributed multi-GPU inference — early and experimental — so larger workloads can be served across many machines, alongside the beginning of open-sourcing the stack.
Phase 3 — Hand it to the community
Progressive decentralization with community-run nodes taking over more of the network — and, if and when it is genuinely justified, a possible $n1o token to coordinate and reward participation. To be clear, this remains a possibility, not a promise.
09 / Open-sourceProgressive decentralization, in the open
Our promise has a mechanism, not just a mission statement. We intend to open the code so the community can inspect it, contribute to it, and depend on it — and then to hand over the network in phases until the community both runs it and earns from it.
"Progressive" is the honest part. Decentralizing everything on day one would mean an unreliable product; keeping everything centralized forever would betray the point. So we decentralize step by step — control of nodes, then coordination, then governance of network parameters — transferring responsibility as the community proves it can carry it. The work happens in the open, so progress against this promise is something you can verify rather than take on faith.
10 / RisksWhat could go wrong
n1o is ambitious and early. You should understand the risks before you use it, run a miner, or form any expectation about the future.
- Experimental technology. The product and network are under active development and may change, break, or be discontinued.
- Distributed inference is unproven. Serving models reliably across many independent consumer GPUs is hard; early distributed features are experimental and may underperform or fail.
- Crypto and regulatory uncertainty. Payments, settlement, and any future token touch a fast-moving, uncertain legal and regulatory landscape that varies by jurisdiction and can change.
- No guaranteed earnings. Miner income varies and is not guaranteed. Do not deploy hardware you cannot afford to run at a loss.
- No guaranteed token. The $n1o token may never exist. There is no price, no sale, and no expectation of profit.
For the full terms, definitions, and legal disclosures, see /legal.