Budget an agent cannot overspend
Escrow is a hard ceiling — the program only releases what a verified action earns, never more than the escrow holds. The guardrail you need before software spends money.
← Use cases · Agentic payments & per-call APIs
A payment leg for the 2026 agent economy: autonomous agents and backends that pay per call, per result, or per conversion against a pre-funded USDC budget — settled on-chain in ~3 seconds, no human approving each transfer.
The agent economy runs on many small, attributable actions — an agent buying compute, calling a paid API, or rewarding another agent for finishing a task. Traditional affiliate and referral accounting cannot settle those economically: a monthly reconciliation cycle and a percentage cut do not fit a workload of tiny, high-frequency, machine-driven events. And you cannot hand software an open-ended credit card.
Njord gives an agent operator a funded escrow the agent cannot overspend, and settles each verified action on-chain for roughly $0.00025 in about three seconds. This is the same escrow-and-settle machinery that pays affiliates — reframed as the pay-per-action leg the x402 / HTTP-402 stablecoin stack is converging on. The multi-party model (funder, agent/earner, optional bridge) maps cleanly onto agent-to-agent and agent-to-service settlement.
Escrow is a hard ceiling — the program only releases what a verified action earns, never more than the escrow holds. The guardrail you need before software spends money.
A typical action settles for ~$0.00025 in ~3 seconds — economical enough to sit inside a request/response loop.
The @njord/sdk submits actions from backends and agents; payout fires on verification, not on invoice approval. @njord/indexer reads history over GraphQL.
More use cases
SaaS & digital products
Creators, publishers, devrel
Bridge & payments operators
The quickstart runs the full lifecycle end-to-end in under an hour.