Live · testnet

Soft Yeti

Donate spare GPU compute to AI inference. Earn YETI tokens automatically.

Double-click the downloaded file — it installs itself to a soft-yeti folder in your home directory and walks you through the rest.

Windows only today · any GPU works, VRAM decides the model · Ollama installs automatically

Active Volunteers
Blocks Minted
YETI Earned by Miners
Checking network activity…
1
Download
One script sets everything up
2
Mine
Your GPU runs real AI tasks
3
Earn
YETI tokens accumulate in your wallet

How it actually works

Most crypto mining burns electricity on math nobody needs. Soft Yeti flips that: the AI inference itself is the "nonce search." Your GPU runs a real task (via Ollama, running locally on your machine) and hashes the output. If the hash happens to meet a difficulty target, you've mined a block — and either way, a real AI task just got answered. Nothing is wasted the way it is in Bitcoin mining: every attempt, successful or not, produces real inference work for the network.

There's no pre-mine and no ICO. Every YETI in existence came from a volunteer's GPU actually doing verified work. YETI also doubles as the subscription currency for the AI product this network powers — you can already spend it at the portal — so mining and spending are two sides of one system, not disconnected token games.

Not every task uses the lottery, though. Work that needs a fast answer — real product requests, not the mining game — can skip the hash search entirely: the first quality-passing result is accepted and paid immediately, at a smaller per-token rate instead of a jackpot. For these instant tasks, what you're paid is computed from the coordinator's own measurement of your output — never from what you claim to have done.

Don't take our word for it

Every block, every reward, and every wallet balance on Soft Yeti is public and independently checkable — not a claim on this page, something you can go verify yourself right now. See exactly which volunteer mined which block, when, and for how much.

Open the Chain Explorer →

Verification, honestly explained

A volunteer network only works if dishonest participants can't cheaply fake the work. Here's what we actually check today — including what we can't verify yet, stated plainly rather than glossed over.

Proof-of-Inference nonce search Live
Skipping real inference and forwarding your task to a cloud API instead is economically irrational once the difficulty target makes that not worth it — the reward math only pays out for volunteers actually running the search locally.
Instant delivery for latency-sensitive work Live
Some real product tasks skip the hash lottery and pay out on the first quality-passing answer instead, trading a smaller per-token reward for zero wasted grinding. The reward is still computed from the coordinator's own measured output — never a volunteer's self-report — and bound to that exact task so a resubmission can't double-pay.
Memory-hard proof of work (Argon2) Live
Closes the cheapest real cheat: grinding fake text through plain SHA-256 on a bare CPU is orders of magnitude cheaper than running real inference. Argon2's memory-hard cost makes forging a result cost about the same as doing the work honestly.
Temperature-zero fingerprinting (canary tasks) Live
About 5% of tasks are "canaries" with a known-exact correct answer, phrased differently each time so the check can't be pattern-matched around. Catches a client that isn't really running inference at all.
Benchmark verification Live
Every submission includes a small computation seeded from that specific task, which we independently recompute and check server-side — a forged or copy-pasted result can't pass. What this does not prove: that the computation ran on a GPU specifically rather than a CPU. No timing measurement can prove that without real hardware attestation, which is out of scope at this network's current size — and it turns out not to matter economically either, since the reward math doesn't reward lying about your hardware tier.
Live · testnet

Why YETI is designed to have real value

Mining is only half the loop. The other half is a real AI product you spend YETI on: sign in with your wallet at the portal, subscribe with the YETI you mined, and describe a project in plain language — J-Claw, our autonomous software factory, builds it and hands you the result. That flow is live on testnet today, so YETI already buys something real instead of the network leaning on token speculation.

Being honest about what's still being built: subscription payments and the network's mint fee both accrue to a treasury that can redistribute to miners — but the automatic payout side, where miners directly earn what subscribers spend, isn't wired up yet. The revenue is real and recoverable; the loop that returns it to miners is the next thing on the roadmap, not a finished feature. We'd rather say that plainly than claim a closed loop we haven't shipped. Progress is tracked openly on the roadmap.

Frequently asked questions

Is this a scam?
The entire coordinator and client are open source on GitHub — read the code yourself rather than take our word for it. There was no pre-mine and no ICO; every YETI token that exists was earned by a volunteer's GPU doing real, verified work.
What does this actually do on my computer?
It runs Ollama locally and answers AI tasks with it. The only thing that leaves your machine is the specific task prompt and your model's output for that one task — nothing else on your computer is read, scanned, or transmitted.
Can I stop anytime?
Yes. The dashboard has an on/off toggle for pausing mining whenever you want, and a real uninstaller (uninstall.ps1, or the "Uninstall" button in the dashboard) that backs your wallet up to your Desktop by default before removing anything — you're never one click away from losing funds by accident.
Is my wallet safe?
Your wallet is encrypted at rest with a random passphrase you never see or type — it's generated automatically and protected by Windows' own per-account encryption (Credential Manager/DPAPI), so only your Windows account on your machine can unlock it.
What happens if the coordinator goes down?
Being straightforward here rather than overselling it: today this is early-stage, centrally-operated infrastructure with a blockchain-style ledger and reward system layered on top — not yet a fully decentralized network. There is automatic failover between two independently-operated machines if the primary goes down, but this isn't the same claim as full decentralization, and we won't pretend it is.
System requirements
Windows only today (Linux/Mac support isn't built yet — said plainly rather than buried). Ollama is required but the setup script installs it automatically if it's missing. Your GPU's VRAM decides which model you'll run:
VRAMModel
< 4 GBqwen2.5:1.5b-instruct
4–6 GBphi4-mini
6–10 GBqwen2.5-coder:7b-instruct
10–20 GBdeepseek-coder-v2:16b
20+ GBqwen2.5-coder:32b
How early is this, really?
Genuinely early — this is a real, live, actively-developed network with real volunteers already mining and real blocks on-chain (check the live numbers above, or browse every block yourself in the chain explorer). You'd be joining as one of the first outside volunteers, not the thousandth. If that sounds interesting rather than risky, this is the right time.
⬇  Download Setup Script

Or clone and run manually (Windows):

git clone https://github.com/Matt28296/soft-yeti cd soft-yeti powershell -ExecutionPolicy Bypass -File setup_volunteer.ps1
Open source on GitHub  ·  Sign in / Portal  ·  Chain Explorer  ·  Roadmap