prism.

Agent-native GPU compute

GPU compute your agents
can rent themselves.

An agent authenticates with its wallet, leases an NVIDIA GPU, and pays per second in USDG. No console, no API keys. Every lease is metered and settled onchain.

Live$0.80 / GPU hourRobinhood Chain · USDG

01 / Network

Compute and settlement
in one service.

Every machine on the network today is rented from a vetted provider and matched to a lease on demand. Operator-owned hardware is not part of the service, and the trust class on each offer says what the host can and cannot see.

R / 01

Rent a GPU

48 GB NVIDIA cards at $0.80 per hour.

Run digest-pinned OCI images in ephemeral workspaces with temporary SSH access. Offers reflect real-time capacity, and a quote names the exact card before you fund anything.

Explore compute ↗
S / 02

Provider program

Operate GPU capacity and earn from settled usage.

The hardware, bonding and isolation requirements for operator-owned NVIDIA hosts are published. No independent host is enrolled today. The settlement terms pay a provider 90% of the confirmed charge on a finalized lease.

Review requirements ↗

02 / Agents

Built for autonomous
agents, not dashboards.

An agent needs three things to use a GPU: an identity, a way to pay, and a way to run code. On Prism the wallet is the identity, USDG is the payment, and an open SDK and MCP server are the interface.

01

Wallet identity

An agent signs a challenge with its wallet to authenticate. No browser login, no API keys to provision or rotate.

02

SDK and MCP

Lease a GPU and run a command in a few lines with the SDK, or connect the MCP server and an agent leases compute as a native tool.

03

Pay per job

Fund a lease in USDG onchain, or settle a single job over x402. An agent pays for exactly the compute it uses.

03 / Settlement

A controlled
lease lifecycle.

01

Quote

Review the per-second rate, funded duration, and maximum USDG commitment.

02

Fund

Approve the quoted amount and deposit it into the lease escrow contract.

03

Provision

Prism assigns an available NVIDIA card and prepares an ephemeral workspace.

04

Meter

Charges begin after GPU and access-readiness checks complete.

05

Settle

Confirmed usage is settled onchain and unused escrow is returned.

04 / Private data

Your secrets stay yours.

Cards, identity documents and credentials go in your vault, sealed on your machine under a key derived from your wallet and never sent. Prism stores the ciphertext and holds no way to read it. Each item names the weakest class of workspace it may ever be shown to, and an agent that asks for more is refused.

Security scope

A rented workspace is a separate matter. Every machine available today can be read by the company hosting it, so new vault items start out of reach of one. Stronger isolation is earned from hardware attestation, never from a supplier’s own claim.

05 / Receipts

Every lease
leaves a receipt.

When a lease settles, Prism writes an onchain receipt: the GPU, the runtime, and the USDG charged. It records the cost, never the contents, so your spend reconciles cleanly and the workspace stays private.

See the receipts ↗
SETTLEMENT RECEIPTONCHAIN
STATUS
SETTLED ONCHAIN
GPU MODEL
NVIDIA RTX 6000 ADA
BILLING
PER SECOND, USDG
NETWORK
ROBINHOOD CHAIN
FEED
/PROOF
Receipts are live

Give your agents a wallet and a GPU.

Launch a workspace.

Find compute ↗Become a provider +