ENEMY 01
Permanent API rent
You build your product on an endpoint whose price, limits, and rules can change the moment it's inconvenient for you.
Alpha available now · live on Koinos testnet
Take AI back from
Take AI back from
the data centers.
A few corporations hoard the GPUs, meter every request, and keep your private data on their servers. Koinos AI puts the first layer back in your hands — private AI on hardware you already own — and turns the rest into a decentralized network no single company controls. The alpha is downloadable today, and the network is live on Koinos testnet (harbinger): chatting locally is free and private, and earning pays KAI on testnet. Two ways to take it back:
Windows 10/11 · free · alpha earnings are testnet KAI — no monetary value yet.
01 · people
Run & earn
Private AI. Idle GPU that pays.
Chat privately on your own machine. Turn unused compute into KAI.
Start earning → 02 · buildersSelf-host
Your API. Your GPUs first.
An OpenAI-compatible endpoint you control that overflows to the network.
Tester quickstart →
OpenAI-compatibleNo KAI to startPrivate by architectureSettled on Koinos testnet
The stakes
Whoever owns the compute owns the intelligence.
Intelligence is becoming the most valuable resource on Earth — and it's being locked inside a handful of hyperscale data centers owned by a few of the largest companies that have ever existed. That isn't a market. It's a monopoly on thought.
Koinos AI is the counter-move: millions of machines, owned by the people who run them, coordinated by an open network no single company controls — with the private machine on your desk as the default, not the exception.
ComputeThe GPUs that produce modern intelligence are hoarded in hyperscale clusters most people can never access.
DataYour prompts, files, and source code become their asset the moment they leave your device.
PowerPrices, rate limits, and who's even allowed in change on their terms — never yours.
The status quo
Modern AI runs on someone else's computer — and bills you for it forever.
The dominant model concentrates the GPUs, the pricing power, and your data inside a handful of data centers. Meanwhile the capable machine on your desk sits idle most of the day.
ENEMY 02
Your data, their servers
Prompts, files, source code, and business context leave hardware you control every time you hit a cloud model.
ENEMY 03
Idle compute, wasted
Millions of capable GPUs do nothing all night while all AI demand is funneled to hyperscale clusters.
The architecture
One request. Four places it can run — you choose which.
Every request is checked against your privacy and spending policy before anything about quality, speed, or price. Local always wins first. The network is a fallback you opt into.
L
LocalRuns on your machine. Strongest privacy.
no KAI chargeP
Private poolHardware your org owns or explicitly trusts.
internalN
Verified networkEncrypted transport to reputation-bearing providers.
settled in KAIC
Confidential · futureAttested confidential-compute hardware.
premium KAIFor developers · your way out
Stop renting your product's brain from a company that competes with you.
Every app built on a hyperscale API is one price hike, rate-limit, or deprecation away from breaking — and it hands your users' data to the same firm racing to ship your product itself. Koinos AI is the exit: self-host an OpenAI-compatible gateway on hardware you own, keep your GPUs as the primary inference layer, and overflow into a decentralized network by policy, not a rewrite. Same SDK. Scoped app keys, never wallet keys. Nobody holding your roadmap hostage.
app.py · drop-in OpenAI SDK
# same SDK. your endpoint. your hardware first. client = OpenAI( base_url="http://localhost:8181/v1", api_key="kai_proj_live_••••", ) r = client.chat.completions.create( model="koinos-smart", messages=[{"role":"user","content":"…"}], # overflow policy, not a different API extra_body={"routing":"local-first", "max_spend_kai":2.5}, )
Local onlywarmNever leaves the box. Zero network charge.
Private-pool firstwarmYour trusted GPUs before any public provider.
Local → network overflowelasticBurst to the public network only when local is saturated.
Lowest cost / fastestelasticOptimize eligible routes under your budget + privacy rules.
Included
Streaming · project keys · per-request/day/month budgets · hard stops · model pinning · usage logs. Embeddings, tools & batch on the roadmap.
For you · earn with your computer
The data centers get paid for compute like yours. Now you do.
The same GPU that runs your games can run real AI work for the network. Install, let it benchmark your hardware, and press one button — Koinos AI handles models, safe resource limits, and pricing. You earn KAI for verified useful work, not idle uptime, and intelligence gets a little less centralized every time you do. During alpha, earnings are testnet KAI on Koinos testnet (harbinger) — they have no monetary value yet.
Estimator · pick your hardware
Illustrative model — a relative earning weight from GPU class × hours × demand, not a yield promise. Real KAI reward depends on verified work, live network demand, and the KAI reference price, all still in economic simulation. Alpha earnings are testnet KAI — no monetary value yet.
—
KAI / day · est.
Eligible workLLM · Embed
Est. KAI / month—
STEP 01
Install & detect
Koinos AI finds your CPU, GPU, VRAM, and drivers and configures a compatible runtime and model automatically.
Download for Windows — free →STEP 02
Press Start Earning
No KAI, KOIN, MANA, or manual pricing. Foreground use always wins; battery and thermal limits protect the hardware.
STEP 03
Do verified work
Completed, verified AI jobs settle in KAI through Koinos — with gasless onboarding sponsored underneath.
The difference, plainly
Centralized AI vs. AI you own.
Same familiar API surface. A completely different relationship to your compute, your data, and your bill.
| Koinos AI | Centralized AI API | |
|---|---|---|
| Where inference runs | Your hardware first, network on demand | Always their data center |
| Your prompts & files | Stay local unless you opt out | Leave your control by default |
| Cost floor | $0 for local — you own the compute | Per-token rent, forever |
| Idle GPU | Earns KAI for verified work | Earns nothing |
| Elastic scale | Overflow into a decentralized network | Only theirs, at their price |
| Lock-in | Open, self-hostable, forkable | Their SDK, their terms |
Privacy by architecture
Your data stops being their asset. You decide where it runs.
To Big AI, your prompts and files leave your control the moment you hit send. Koinos AI flips the default: local execution never leaves your device, and when you do reach for the network you pick the exact privacy mode — and see precisely where your request is allowed to run.
Where this request may run
Localyour machine
→
Privatetrusted pool
→
Networkpublic providers
→
Confidentialattested TEE
The KAI economy
Value flows to useful work — not to a meter on your own machine.
KAI settles the network: providers earn it for verified compute, model creators can earn bounded royalties, and you spend it only when you use resources someone else supplied. Local inference has no KAI charge, ever.
Share compute → earn KAI
Benchmark, route, protect foreground use, and settle verified AI work automatically. Providers can earn KAI without first buying KAI or KOIN.
Native rewardKAI
Secure Koinos → earn KOIN
An optional module runs a full node and, if you choose, Proof-of-Burn block production. A separate earning path with its own economics — no GPU mining required.
Native rewardKOIN
Clean by design: KAI rewards useful AI work. KOIN rewards Koinos consensus. We never double-pay KAI just for producing blocks. Token supply, fees, and bootstrap pricing stay provisional pending economic simulation and legal review.
Roadmap
From private local AI to distributed intelligence.
Prove the core loop first — install, run local AI, earn, settle, overflow — then earn the harder layers. Gated by readiness, not a calendar.
ERA I
Network Genesis
- Local AI & chat
- Provider earning
- KAI settlement
- Self-hosted API
- Network overflow
ERA II
Open Compute
- Public providers
- More models & hardware
- Private pools
- Batch / spot compute
- Model marketplace
ERA III
Autonomous AI
- Persistent agents
- Tools & permissions
- Machine payments
- Confidential compute
ERA IV
Distributed Intelligence
- Synthetic data
- Fine-tuning
- Distillation
- Distributed training
- Koinos-native models
Early access
Own your AI. Share the scale.
Be first to run private AI on your own hardware — and first to earn from it when the network opens.
✓ You're on the list. We'll reach out as spots open.
No spam. No token sale. A working product first — then price discovery. We keep your email, which option you picked and the time — nothing else: privacy.
FAQ
Straight answers.
Do I need cryptocurrency to use it?
No. Local AI runs on your own hardware with no KAI required. Network services can be paid through familiar payment and credit experiences, with Koinos handling settlement underneath.
Do I need KAI or KOIN to start earning?
No. Entry-level providers start with compatible hardware and sponsored (gasless) Koinos resource usage. Higher-value provider classes may later use KAI collateral — but you never have to buy KAI to begin earning KAI.
Is network inference private?
Not automatically — and we won't pretend otherwise. Local mode never leaves your device; private pools restrict execution to trusted hardware; verified-network mode uses encrypted transport; confidential compute (future) adds hardware attestation. You choose the fallback order.
Why does the self-hosted API matter?
It keeps your own GPUs as the primary inference layer while giving you an elastic path to decentralized capacity — instead of building your product permanently around a hyperscale API whose terms you don't control.
Is this just crypto mining?
No. You're not rewarded for burning power on hashes — you earn KAI for completing verified, useful AI work: inference, embeddings, evaluation, and later fine-tuning. Passive uptime isn't the reward.