Skip to main content

Coding Agents Configuration

This section explains how to configure various AI coding agents to work with the PwC AI-CoE Coding Agents Gateway.

API Compatibility

The gateway supports two types of API compatibility, depending on the coding agent you're using:

API TypeCoding AgentsAuthentication Method
Anthropic Messages APIClaude CodeEnvironment variables in settings.json
OpenAI Chat Completions APICline, Codex, OpenCodeCustom headers

Before You Start

Before configuring any coding agent, make sure you have:

  1. ✅ Obtained your API key and tenant ID (see Get Access)
  2. ✅ Reviewed your budget limits (see Budget)
  3. ✅ Chosen a model appropriate for your agent (see Available Models)

Supported Coding Agents

Select your coding agent to view the configuration guide:

  • Claude Code - Anthropic's official CLI coding agent
  • Codex - OpenAI's coding agent
  • Cline - VS Code extension with multi-provider support

Interactive API Reference

You can explore and test all gateway endpoints directly in your browser — no client setup required.

ToolURLNotes
Swagger UIhttps://idi-coding-agents.pwc.it/docsTry requests interactively
ReDochttps://idi-coding-agents.pwc.it/redocClean reference view

Both are served by the gateway itself and require VPN access.

To authenticate in Swagger UI, click Authorize and paste your credentials in one of the two supported formats:

  • BearerAuth — paste apikey=<your-api-key>&tenantid=<your-tenant-id> into the Bearer token field
  • ApiKeyHeader + TenantIdHeader — fill in the api-key and tenant-id fields separately

Network Requirements

All coding agents require access to the gateway endpoint:

EndpointNetworkNotes
https://idi-coding-agents.pwc.itPwC VPN (Italy or Italy On-Premise)Recommended
https://doc-interact-backend.pwc.itPublic internetNot recommended - may be blocked by Imperva firewall
warning

The public endpoint (doc-interact-backend.pwc.it) is protected by the Imperva firewall, which may block requests from coding agents due to their high frequency, interpreting them as potential DDoS attacks. Always prefer the VPN endpoint when possible.