Guides

Architecture at a glance

A short tour of how AgentRail is put together — enough to make integration decisions, not so much that you have to understand every internal.

Three layers

AgentRail is built in three coordinated layers. You only ever interact with the top one.

  • ─Surface — REST API, MCP server, SDKs, Console, and AgentRail Chat. This is what you call.
  • ─Platform — identity, policy, ledger, observability, compliance, dispute lifecycle. Shared across products.
  • ─Rails — direct integrations with banks, card networks, and payment systems. Hidden behind the platform layer.
Note
Most teams never need to think below the Surface layer. The Platform is what powers the things we promise (policy enforcement, audit, real-time settlement) and the Rails are the implementation detail that makes them happen.

Trust boundaries

Every action crosses a series of trust boundaries before money moves: your agent calls our API, our API verifies your credential, the platform enforces your policy, compliance screens the counterparty, and only then is a rail engaged.

Each boundary fails fast. A policy violation at boundary three never becomes a rail debit at boundary five — you can't accidentally move money you weren't authorized to move.

Rail abstraction

AgentRail exposes one payment object. Behind it, we route across ACH, RTP, FedNow, wires, and card payouts based on speed, cost, and availability for the recipient. The selected rail is recorded on every payment so you have full visibility — you just don't have to choose.

← Back to docs indexAsk AgentRail Chat →