Architecture Overview

Architecture Overview

If you've integrated multiple payment providers, you know the pain:

  • Stripe uses PaymentIntents

  • Adyen uses payments

  • PayPal uses orders

All of them do the same job, but each has different field names, different status enums, different error formats.

This problem exists in other domains too, but solved with well maintained developer centric libraries, open source and free from vendor lock-in.

Domain
Unified Interface
What It Solves

LLMs

One interface for OpenAI, Anthropic, Google, etc.

Databases

One ORM for PostgreSQL, MySQL, MongoDB, etc.

Cloud Storage

One CLI for S3, GCS, Azure Blob, etc.

But for payments, no such equivalent exists for developers.

Prism is the unified abstraction layer for payment processors—giving you one API, one set of types, and one mental model for 100+ payment connectors.

Architecture Components

The Prism supports a three layered architecture, each solving a purpose. The architecture prioritizes:

  1. Consistency: Same types, patterns, and errors across all connectors

  2. Extensibility: Add connectors without SDK changes

  3. Developer Experience: Idiomatic payments interface with multi language SDKs

Component Descriptions

Component
Problem It Solves
Technologies

Interface Layer

Developers can think in their language's patterns while using the unified payments grammar. You use client.payments.authorize() with idiomatic types in your codebase

Node.js, Python, Java, .NET, Go, Haskell

Binding Layer

Each language needs native-performance gRPC with seamless transport without language bridges; handles serialization

tonic, grpcio, grpc-dotnet, go-grpc

Core Layer

Single source of truth for payment logic with freedom to use Prism as a separate microservice. One implementation serves all languages; also include connector adapters maintaining the request response mapping to 100+ processors from the Proto

Rust, tonic, protocol buffers

Data Flow

spinner

Connector Transformation

The core value of the Prism is transformation from a single unified interface into multiple processor patterns. For easier understanding, a simple example of how a Stripe Authorize Request and an Adyen Authorize Request is mapped against the Unified interface.

Authorization Mapping:

Unified Field
Stripe Request
Adyen Request

amount.currency

currency

amount.currency

amount.amount

amount (cents)

value (cents)

payment_method.card.card_number

payment_method[card][number]

paymentMethod[number]

connector_metadata

metadata

additionalData

This transformation happens server-side, so SDKs remain unchanged when adding new connectors.

Connector Adapter Pattern

Adding new connectors into PRism should also be easy and declarative. It is simplified with a standard interface for the ConnectorAdapter trait.

Adding new connectors only need an adapter implementation. SDKs require zero changes.

Last updated

Was this helpful?