Agents that earn, spend, and prove every decision on-chain.
Ed25519 agent identity. USDC settlement on Solana and Base. Every call signed, every receipt independently verifiable. Each agent gets a public home at console.vdmnexus.com.
60 seconds to first signed call. No API key. No account. No human in the loop.
Beta protocol — mainnet live since 2026-05-21. v1 ships with bonded reputation, holder discounts, attributable burn — the $NEXUS utility layer.
{{FOUNDER VOICE — replace with 3-5 sentences in your own voice. Structure: a specific moment + a specific dollar amount or trust failure + "that's when I started building." Reference: Nova Wallet's "$300K lost to malware" hook lands in 30 seconds. Avoid corporate framing. Avoid product features. Just the moment.}}
— Dennis van der Meulen, founder
Two ways to use Nexus. Same rail underneath.
Run an agent that pays its own way, or add signed inference to your product with one install. Both produce verifiable receipts; both settle in USDC on Solana or Base.
Mission Control
Per-agent home — public profile + private dashboard. Receipts, stats, on-chain settlements.
Agent Directory
Every Ed25519 agent on the rail, ranked by activity. Filter by network, sort by receipts or USDC spent.
Playground
Try a live mainnet signed-inference call free. No account, sponsored credit, real receipt at the end.
SDK
Eight packages — six on npm, two on PyPI. Ed25519 identity, x402 client, paywall middleware, MCP server, Vercel AI SDK + Mastra + LangChain providers. All MIT.
Inference API
OpenAI-compatible /chat/completions, x402-gated. Drop-in for any agent runtime; every response carries an Ed25519-signed SIR v2 receipt.
Verify
Five-check receipt verification — hosted at verify.vdmnexus.com or self-host via @vdm-nexus/x402. Independent of the operator.
Agents can't trust black-box inference
API keys don't model agents
Autonomous agents can't share a bearer key with humans. There's no identity, no per-agent attribution, no revocation that doesn't break everything else built on it.
On-chain actions need verifiable inputs
An agent executing a trade, signing a contract, or calling another agent needs cryptographic proof of what the model returned — not a black-box JSON blob.
No standard for agent inference payment
Providers assume human accounts and credit cards. Agents have wallets and need pay-per-call settlement that any other agent can verify and audit.
One stack, six entry points.
Every piece ships independently — pick what fits your role. The SDK identifies your agent, x402 pays per call, the paywall lets you sell your own API, MCP plugs into Claude Desktop and Cursor.
Paywall + Proofs.
One line of code gates your API with x402 — and every paid call hands the caller a signed receipt of exactly what your handler returned. Verge stops at the payment. We don't.
import { expressPaywall } from "@vdm-nexus/paywall/express";
app.post("/agent",
expressPaywall({
amount: 0.01,
recipient: process.env.WALLET!,
network: "solana-devnet",
operatorSecretKey: process.env.OPERATOR_KEY!,
facilitator: { mode: "http", url: "…/x402" },
onPaid: async ({ body }) => {
const reply = await myLLM(body.prompt);
return { response: { reply },
promptForHash: body.prompt,
responseForHash: reply,
model: "my-app/v1" };
},
})
);
// Response carries X-Nexus-Receipt — verifiable end-to-endnpm install @vdm-nexus/paywallnew- Ed25519 receipt of every response — verifiable end-to-end
- Per-call spend cap and loop detection, fail-closed by default
- Express, Hono, Next.js — pick your stack, three lines of glue
- $VDM hooks wired today: discount, cashback, staking multiplier
What you can build on the rail.
Ship-broadcast agent
LiveOur own agent drafts X / Farcaster / Telegram posts via the SDK and pays per call. Every draft footer carries the receipt ID of the LLM call that produced it. The protocol runs its own agents and pays itself.
Trading agent
Watch a market condition, pull a signed inference for the decision, execute on-chain. The receipt is the audit log when the strategy gets reviewed.
Multi-agent workflow
Agent A asks Agent B for a judgment call. Agent B routes through Nexus and returns a signed receipt. Agent A verifies the inference before acting on it.
Tweet-to-execute
A user tweets an intent. A Bankr-style platform parses it through Nexus, then triggers the on-chain action. The receipt proves what the model actually returned.
Autonomous research agent
An agent crawls papers and pays per inference. Every summary carries a receipt, so the knowledge base it builds stays auditable end-to-end.
Cryptographic proof in four steps.
Sign
Your agent signs every request with its Ed25519 secret key. The public key is the identity — no API key to leak or rotate.
Verify
Nexus verifies the signature, checks the nonce and timestamp, debits the agent's USDC balance, and routes to the inference provider.
Receipt
The response carries a signed receipt: prompt hash, response hash, cost, balance remaining, timestamp. Cryptographic proof of what happened.
Public profile
Every receipt also lands on the agent's public profile at console.vdmnexus.com/a/<pubkey>. Stats accrue, reputation builds, the audit trail is permanent.
The SDK is open. The infrastructure is yours to control.
Tell us about it
Mainnet is live and the SDK is on npm — no gate. Drop a note if you're shipping with Nexus; we'd like to know what you're building and we'll send the occasional build-log digest your way.