Monetize Your Compute.
Turn an idle GPU or a spare API key into a high-yield asset. Stake HBAR, register on Hedera, and earn for every request the exchange routes your way.
How Providers Earn
List your compute
- check_circleRun the provider service on any box — VPS, cloud, or your laptop behind a tunnel.
- check_circlePoint it at a Groq API key, or run without one for canned demo answers.
- check_circleAdvertise the model you serve and a price in USDC per request.
Stake & register
- check_circleBoot stakes 50 ℏ into the escrow account as a quality bond.
- check_circleRegistration is posted automatically to the HCS registry topic.
- check_circleThe exchange discovers you from Mirror Node and adds you to routing.
Serve & earn
- check_circleCheapest live provider for a model wins the request.
- check_circlex402 settles the payment in USDC before you return the completion.
- check_circleEvery settlement is an on-chain Hedera transaction — no invoicing.
What You Need
Any box — no GPU required. A Groq API key serves real models; without one, canned answers keep you live.
50 ℏ staked to escrow — slashed if you serve a cheaper model than you advertise.
A testnet account ID + key per provider. The service registers your HCS-14 identity on boot.
Set PROVIDER_PUBLIC_URL to a tunnel or VPS so the exchange can route requests to you.
Quickstart.
The terminal on the right is the whole onboarding. Clone, add your Hedera keys, make your box reachable, and start — the service stakes and registers itself.
git clone github.com/GigaHierz/ETH-global-lisbon-2026 && pnpm install
HEDERA_PROVIDER_ID=0.0.xxxxx · HEDERA_PROVIDER_KEY=0x…
PROVIDER_NAME=Acme · PROVIDER_MODEL=llama-3.3-70b-versatile · PROVIDER_PRICE=0.10
PROVIDER_PUBLIC_URL=https://your-tunnel-or-vps
pnpm provider
curl -X POST localhost:4025/v1/chat/completions \
-d '{"model":"llama-3.3-70b-versatile",…}'
scheme: exact · network: hedera:testnet
amount: 10000000 tinybars · payTo: 0.0.9744152
Configuration
| Variable | Default | Purpose |
|---|---|---|
| PROVIDER_NAME | Custom Provider | Display name shown in the routing table. |
| PROVIDER_MODEL | llama-3.3-70b-versatile | The model you advertise — and must actually serve. |
| PROVIDER_PRICE | 0.10 | Your price per request, in USDC. |
| HEDERA_PROVIDER_ID / _KEY | from pnpm setup-hedera | Account that stakes, registers, and receives payment. |
| PROVIDER_PUBLIC_URL | http://localhost:4025 | Public address the exchange routes to. |
| GROQ_API_KEY | — | Upstream inference. Omitted → canned fallback answers. |
| STAKE_HBAR | 50 | Boot-time stake posted to escrow. |
Provider Endpoints
| Method | Path | Returns |
|---|---|---|
| GET | /info | name, model, price, wallet, agentId, url |
| GET | /healthz | liveness probe used by the exchange |
| POST | /v1/chat/completions | 402-gated inference — pay in USDC, receive the completion |
The verifier replays sampled prompts against a witness provider. Serve a cheaper model than you advertise and the divergence is caught — your 50 ℏ stake is seized to the treasury and you're dropped from routing.
Ready to list your compute?
Join the provider waitlist, or clone the repo and run pnpm provider right now.