JD.SemanticKernel.Connectors.OpenAICodex

OpenAI Codex session connector for Semantic Kernel. Provides UseCodexChatCompletion() to wire local Codex CLI OAuth credentials (or an explicit API key) into any Semantic Kernel application without managing tokens manually.

Features

• Multi-source credential resolution — Options → OPENAI_API_KEY → CODEX_TOKEN → ~/.codex/auth.json → interactive device code login
• Token exchange — Automatically exchanges Codex OAuth tokens for OpenAI API keys via the standard token exchange endpoint
• Token refresh — Automatically refreshes expired tokens using stored refresh tokens
• Device code login — Full interactive codex login equivalent for .NET applications
• Semantic Kernel integration — One-call UseCodexChatCompletion() extension method
• DI support — AddCodexAuthentication() for dependency injection
• Model discovery — IModelDiscoveryProvider with known Codex models (o3, o4-mini, codex-mini, GPT-4.1, etc.)
• Multi-TFM — Targets netstandard2.0, net8.0, and net10.0

Installation

dotnet add package JD.SemanticKernel.Connectors.OpenAICodex

Quick Start

Minimal (auto-discovers credentials)

using JD.SemanticKernel.Connectors.OpenAICodex;
using Microsoft.SemanticKernel;

var kernel = Kernel.CreateBuilder()
    .UseCodexChatCompletion()
    .Build();

With explicit API key

var kernel = Kernel.CreateBuilder()
    .UseCodexChatCompletion(apiKey: "sk-your-openai-api-key")
    .Build();

With specific model

var kernel = Kernel.CreateBuilder()
    .UseCodexChatCompletion(CodexModels.O3)
    .Build();

Using Dependency Injection

services.AddCodexAuthentication(o => o.ApiKey = "sk-...");

// Or from configuration:
services.AddCodexAuthentication(configuration);
// Reads from "CodexSession" section in appsettings.json

Direct HttpClient usage

using var client = CodexHttpClientFactory.Create();
// Use with any OpenAI-compatible API

Credential Resolution Order

The provider checks these sources in order and uses the first valid credential found:

Priority	Source	Description
1	CodexSessionOptions.ApiKey	Explicit API key from options/config
2	CodexSessionOptions.AccessToken	Explicit OAuth token from options/config
3	OPENAI_API_KEY env var	Standard OpenAI environment variable
4	CODEX_TOKEN env var	Codex-specific environment variable
5	~/.codex/auth.json	Codex CLI local credentials file
6	Device code login	Interactive OAuth (when EnableInteractiveLogin = true)

Configuration

appsettings.json

{
  "CodexSession": {
    "ApiKey": null,
    "Issuer": "https://auth.openai.com",
    "ApiBaseUrl": "https://api.openai.com/v1",
    "AutoRefreshTokens": true,
    "EnableInteractiveLogin": false
  }
}

Available Options

Option	Type	Default	Description
ApiKey	string?	null	Explicit OpenAI API key
AccessToken	string?	null	Explicit OAuth access token
CredentialsPath	string?	null	Custom path to auth.json
Issuer	string	https://auth.openai.com	OAuth issuer URL
ApiBaseUrl	string	https://api.openai.com/v1	OpenAI API base URL
ClientId	string?	null	OAuth client ID for device code flow
EnableInteractiveLogin	bool	false	Enable interactive device code login
AutoRefreshTokens	bool	true	Auto-refresh expired tokens

Available Models

Constant	Model ID	Description
CodexModels.O3	o3	Most capable reasoning model
CodexModels.O4Mini	o4-mini	Fast, cost-effective reasoning
CodexModels.CodexMini	codex-mini-latest	Optimised for code generation
CodexModels.Gpt4Point1	gpt-4.1	High-capability general model
CodexModels.Gpt4Point1Mini	gpt-4.1-mini	Fast general model
CodexModels.Gpt4Point1Nano	gpt-4.1-nano	Fastest, most cost-effective

Related Packages

• JD.SemanticKernel.Connectors.ClaudeCode — Claude Code connector
• JD.SemanticKernel.Connectors.GitHubCopilot — GitHub Copilot connector
• JD.SemanticKernel.Connectors.Abstractions — Shared abstractions

License

MIT