The x402 payment standard has multiple diverging facilitator implementations (CDP, Foundation, self-hosted) that do not align perfectly. A payment valid against one facilitator can fail silently against another — and for an AI agent, a failed payment is a lost call and lost funds. The NEAR ecosystem lacks production-grade middleware that absorbs this cross-facilitator divergence and handles native NEAR token payments.

Issue x402-foundation/x402#2437 documents CDP's hosted facilitator rejecting valid upto settlement payloads (setSettlementOverrides) — the mechanism CDP's own seller quickstart prescribes — while the Foundation reference implementation accepts the identical wire shape cleanly. A facilitator-level divergence, confirmed with an on-chain tx and source-traced through the SDK:
https://github.com/x402-foundation/x402/issues/2437

This is one documented instance of a structural problem, not an isolated bug. Interouter is the boundary layer that absorbs that class of divergence.

Implementation (public, verifiable): github.com/decision3-ai/interouter — v0.1.7, 59/59 tests passing, MIT. 5-stage ChainAdapter lifecycle (readState → preparePayment → sign → submit → awaitFinality), budget circuit breaker, structured cross-facilitator fallback. NEAR adapter read-only by design while I confirm settlement wire formats; funding makes NEAR payments transactional (NEP-141 + full NEAR AI-agent flows).

Don't take it on my word — clone it, run the tests, read the adapters. Trust is only earned in the open.

AI agents on NEAR cannot reliably pay for services across facilitators. Every failed payment = lost inference call = real money lost.

Interouter uses a facilitator-agnostic ChainAdapter pattern: each chain/facilitator is a swappable adapter exposing the same 5-stage lifecycle (readState → preparePayment → sign → submit → awaitFinality). A router runs reads in parallel, then a sequential payment fallback — first successful settlement wins — with a budget circuit breaker that prevents overspend and structured error capture, so a single facilitator failure never breaks the call.

For NEAR: extend the existing read-only NearAdapter to transactional, add NEP-141 (FT) token support, and implement the full NEAR AI-agent payment flow against a confirmed settlement wire format. Estimated scope ~3–4 months part-time, milestone-based: transactional NEAR adapter → first live NEAR payment through Interouter → developer docs + reference example.

Interouter v0.1.7 — already implemented. Needs funding to add FT token support (KAIC/NEP-141) and full NEAR AI agent payment flows.