Nori - Pay-As-You-Go Services for Metaplex Agents

Nori is a pay-as-you-go service agent operated by the Metaplex Foundation. It sells LLM inference, image generation, and Solana RPC access to other agents — priced in USD, settled in SOL per call against the calling agent's onchain wallet. Nori is also the open-source reference implementation of a Metaplex service agent: agent builders can study (and copy) its A2A surface, delegate-pay billing, x402 fallback, and rate-card patterns.

Summary

Nori removes the plumbing every agent operator otherwise wires up themselves — LLM provider keys, an image-generation account, paid Solana RPC, and per-call billing. A consumer agent needs only a Solana keypair and a registered agent asset.

• Three metered services — chat.completion (Anthropic / OpenAI / Google, tool calls supported), image.generation (gpt-image-1), and solana.rpc (RPC + DAS pass-through)
• Two payment rails — delegate-pay (primary, one-time onchain setup) and x402 v2 (fallback, per-call HTTP 402 flow)
• Charge-on-success billing — the upstream call runs first; failed calls are never charged, and every charge carries an onchain Memo receipt
• Single point of failure caveat — a delegated agent depends on Nori's availability for inference, images, and RPC; see Nori as a single point of failure for mitigations

Services Nori Provides

Nori exposes three services over two surfaces that share one handler stack. Skill input/output uses canonical OpenAI wire format for chat and images, and standard Solana JSON-RPC for RPC — an A2A caller and an OpenAI-SDK caller send byte-identical payloads.

Both surfaces reach the same services:

• OpenAI-compatible HTTP (/v1/*) — point any OpenAI SDK or AI framework at Nori with baseURL. This is the surface most consumer agents use.
• A2A JSON-RPC (/a2a) — programmatic agent-to-agent calls. Discovery starts at GET /.well-known/agent-card.json, which advertises skills, payment schemes, and Nori's serviceExecutiveAddress (the address you register as a delegate).

How Nori Payments Work

Nori picks a payment rail per call: delegate-pay when the caller has onboarded, x402 otherwise.

The delegate-pay rail is what makes Nori invisible to your agent's end users: after a one-time delegation, every call settles automatically with no wallet prompts and no over-quoted holds. Pricing is published on a versioned rate card with a price-change notice policy.

Nori as a Single Point of Failure

A delegated agent that sources its inference, image generation, and RPC from Nori has made Nori a single point of failure: if Nori is unavailable, the agent loses those capabilities until Nori recovers. This is the top-ranked risk in Nori's own risk register, and the v1 mitigation is documentation and portable interfaces rather than redundancy.

Plan for it explicitly:

• Interfaces are portable by design. chat.completion is canonical OpenAI wire format and solana.rpc is standard Solana JSON-RPC. A break-glass fallback is a config change: point your OpenAI-compatible client at another provider (with your own key) and your RPC calls at any public or paid endpoint.
• Keep break-glass credentials. Zero-BYOK is Nori's convenience, not a requirement of your architecture. Holding a low-tier provider key and a free RPC URL in reserve keeps your agent degraded-but-alive during a Nori outage.
• The x402 rail is an independent fallback for payment, not availability. It removes the delegation dependency but still depends on Nori being up.
• Delegation is revocable at any time. If you migrate off Nori, the asset owner revokes the delegation record and the delegate-pay rail hard-stops.

Using Nori as a Reference Implementation

Nori is the working blueprint for a Metaplex service agent — an agent that charges other agents for work. The source repository demonstrates each pattern end-to-end:

To add a new paid service in your own fork: write a payment-agnostic handler that returns a result plus costUsd, add pricing to the pricebook, wire it into the A2A skill dispatch, and declare it on the agent card.

Quick Reference

Notes

• Nori's deployed base URL is published via its agent registration; examples across this section use NORI_URL as a placeholder for the base URL
• Charges are priced in USD and converted to SOL at charge time using the live Jupiter SOL/USD price (30-second cache) — see Pricing and Billing
• Delegation grants Nori billing authority over your agent's PDA wallet. Keep only a working balance there and audit the Memo receipts on each charge
• message/sendStream is declared on the agent card but returns 501 in v1; A2A calls are synchronous
• Nori (the hosted Metaplex service) and agent-plumber (the open-source implementation) are the same codebase; this documentation uses "Nori" for both

Maintained by Metaplex Foundation. Last verified: 2026-07-08.

FAQ

Common questions about Nori.

What is Nori?

Nori is a pay-as-you-go service agent operated by the Metaplex Foundation. It sells LLM inference, image generation, and Solana RPC access to other agents, priced in USD and settled in SOL per call against the calling agent's onchain PDA wallet. It is also the open-source reference implementation for building Metaplex service agents.

Do I need my own LLM provider API keys to use Nori?

No. Nori holds the upstream provider keys (Anthropic, OpenAI, Google, image generation, paid Solana RPC). A consumer agent only needs a Solana keypair and a registered agent asset — every call settles per-use in SOL against the agent's PDA wallet.

What happens to my agent if Nori goes down?

A delegated agent that relies on Nori for inference, images, or RPC loses those capabilities while Nori is unavailable. Nori's surfaces are OpenAI-compatible and standard Solana JSON-RPC, so a break-glass fallback is pointing your client at any other OpenAI-compatible provider or RPC endpoint with your own keys. Self-hosting your own Nori instance is planned for v2.

Is delegating to Nori safe? Can Nori drain my wallet?

Delegation grants Nori billing authority over your agent's PDA wallet, so only keep a working balance there. Every charge carries an onchain Memo receipt you can audit, charges are only taken for successful calls, and the asset owner can revoke the delegation at any time, which hard-stops the delegate-pay rail.

What is the difference between the delegate-pay rail and the x402 rail?

Delegate-pay is the primary rail — after a one-time onchain delegation, Nori charges your agent's PDA directly per call with no payment round-trip. x402 is the fallback for non-delegated callers — the first request returns HTTP 402 with payment requirements, the caller pays (SOL or USDC), then retries.

Glossary

Core terms used across the Nori documentation.