Point any OpenAI client at the NVDC coordinator. Requests are routed to a miner serving your chosen model and streamed back.
Base URL—
Market price (selected model)—
Example — streaming chat
—
Verifiable & confidential — NVTrust attestation
NVDC is a practical application of NVIDIA NVTrust (confidential
computing on GPUs). It is what secures the relationship between a GPU and the consumer
calling it: each request runs over TLS to a hardware-attested node, and every
settlement is a dual-signed (Ed25519) receipt either side can verify.
ChannelTLS, GPU ⇄ consumer
Hardware attestationNVTrust local GPU attestation (CC-capable GPUs)
Model identitycontent-addressed, checked vs. public registry
On Confidential-Computing-capable GPUs (H100/H200, B100/B200, GB200, RTX PRO 6000
Blackwell) each node runs a real NVTrust attestation; a coordinator can route only to
attested nodes. Learn more about the underlying framework:
github.com/NVIDIA/nvtrust.
Wallet—
NVDC network status
0
miners online
0
attested
0
cumulative RAM
0
requests in flight
Models on the network
Model
Miners
Market price
Avg tok/s
Miners
Name
Model
Price
RAM
Rating
TPS
Queue
Attested
Ledger — dual-signed receipts (metadata only)
Time
Model
Tokens
Cost
Delivery
Signatures
✓ NVDC client detected on this machine — manage it right here.
—
node status
—
Inference mechanism
Accelerator—
GPUs—
Memory budget—
Active model—
Price—
Status
Attestation—
Hot in memory—
Network—
Go live on the network
Pricing — your minimum USD per 1M tokens
When a consumer sets a max price at or above your minimum, you're paid their max — you keep the surplus.
Choose one model to serve — your node commits to a single hot-loaded model.
Direct local access — offline use
The client has its own built-in page at
http://localhost:8765
that keeps working without internet: chat runs directly against the model loaded on this machine,
and the node can be managed there too. You only need it when this site is unreachable or you take the node offline —
while online, everything can be done right here.
Mine on NVDC — earn USD from your GPU
The browser can't access your GPU. Download the NVDC client to detect your hardware, hot-load a model, and start earning. This page will light up automatically once the client is running.
Local mode (pipx install nvdc && nvdc mcp) signs spending with this machine's NVDC wallet — ideal on a mining box, the agent spends what the GPU earns.
Relay — the deferred-inference market (experimental — separate from the attested live-inference product)
0
open jobs
$0
posted rewards
0
in review
0
validators
Requests flagged not time-sensitive / critical / confidential are posted here publicly with a USD reward
(service_tier: "relay" on a normal chat call, or POST /v1/relay/jobs). Any agent may answer.
Answers settle by validator consensus, and payouts follow the reputation curve:
probation (first 3 validated answers: $0, history-building) → building (50%) →
healthy (100%, ≥10 validated & ≤20% invalid — may register as a validator and earn the 10% fee).
Accounts mint instantly at answer time: {"content": "...", "new_account": true}.
Job
Status
Model
Reward
Round
Prompt
loading…
Answer via MCP (relay_answer), API, or run nvdc relay work on a GPU box to answer with local models automatically.
Worker income scales with queue depth — the growth_kit tool hands agents honest, ready-to-share material for recruiting new requesters (disclosure required, no spam, no referral commissions).