Skip to main content

Introduction

The PwC AI-CoE Coding Agents Gateway is a service deployed under the IDI (Instant Document Interaction) GenAI Platform architecture. It provides an abstraction over PwC's GenAI Shared Service, offering access to the same LLMs but with centralized governance under the IDI infrastructure.

Key Features​

  • Multi-Provider Support - Access models from Google (Gemini), Anthropic (Claude), OpenAI (GPT), Azure OpenAI, and PwC self-hosted models through a single gateway.
  • Automatic Model Selection - Send model: "auto" and the gateway classifies your request and routes it to the most appropriate model.
  • OpenAI API Compatibility - Use OpenAI compatible Coding Agents (e.g. Codex, Cline, OpenCode).
  • Anthropic Messages API Compatibility - Use Anthropics compatible Coding Agents (e.g. Claude Code).
  • Budget Management - The same built-in budget management system of IDI.
  • Individual API Keys - Each user gets its own api key with dedicated budget and usage monitoring.
  • Interactive API Reference - Explore all endpoints and try calls directly in your browser via the built-in Swagger UI (/docs) or ReDoc (/redoc).

Architecture​

The gateway acts as a unified entry point for AI coding agents, handling authentication and quota enforcement before routing requests to PwC's GenAI Shared Service.

┌─────────────────┐      ┌─────────────────────────────────────────┐      ┌─────────────────────┐      ┌─────────────────┐
│                 │      │   PwC AI-CoE Coding Agents Gateway      │      │                     │      │  LLM Providers  │
│  Coding Agent   │ ───▶ │                                         │ ───▶ │  PwC GenAI Shared   │ ───▶ ├─────────────────┤
│                 │      │  ┌──────────────┐  ┌──────────────────┐ │      │      Service        │      │ OpenAI          │
│ Claude Code     │      │  │ Per-User     │  │    Budget        │ │      │      Lite LLM       |      │ Azure OpenAI    │
│ Cline, Codex    │      │  │ Auth         │  │ Management       │ │      │                     │      │ Google VertexAI │
│ ...             │      │  └──────────────┘  └──────────────────┘ │      │                     │      │ Amazon Bedrock  │
└─────────────────┘      └─────────────────────────────────────────┘      └─────────────────────┘      └─────────────────┘

Request Flow​

  1. Coding Agent sends a request with user credentials (API key + tenant ID).
  2. IDI Authentication Service validates API key and tenant-id to authenticate and authorize the user.
  3. Budget Management System checks and enforces usage limits.
  4. The gateway routes the request: cloud models go to the PwC GenAI Shared Service (which forwards to the appropriate LLM provider); self-hosted models go to the PwC Agent orchestrator on PwC infrastructure.
  5. Response streams back through the same path to the coding agent.

Next Steps​