JD.SemanticKernel.Connectors.GitHubCopilot 0.1.7

JD.SemanticKernel.Connectors.GitHubCopilot

Use your GitHub Copilot subscription as an AI backend for Microsoft Semantic Kernel applications.

This connector reads your local Copilot OAuth credentials, exchanges them for short-lived API tokens, and injects authentication into SK's OpenAI-compatible chat completion — giving you access to GPT-4o, Claude, Gemini, and more through a single Copilot subscription.

Quick Start

dotnet add package JD.SemanticKernel.Connectors.GitHubCopilot

using JD.SemanticKernel.Connectors.GitHubCopilot;
using Microsoft.SemanticKernel;

// One-liner: reads credentials from your local Copilot installation
var kernel = Kernel.CreateBuilder()
    .UseCopilotChatCompletion()
    .Build();

// Use any model available through Copilot
var kernel = Kernel.CreateBuilder()
    .UseCopilotChatCompletion(CopilotModels.ClaudeSonnet4)
    .Build();

How It Works

┌──────────────────────┐     ┌─────────────────┐     ┌──────────────────────┐
│  Local Copilot Auth  │     │  Token Exchange  │     │    Copilot API       │
│  apps.json/hosts.json│────▶│  api.github.com  │────▶│  chat/completions    │
│  (ghu_* OAuth token) │     │  /copilot_internal│    │  (OpenAI-compatible) │
└──────────────────────┘     │  /v2/token       │     └──────────────────────┘
                             └─────────────────┘
                              ▲ Short-lived token
                              │ (~30 min TTL)
                              │ Auto-refreshed

1. Reads your local Copilot OAuth token from %LOCALAPPDATA%/github-copilot/apps.json (Windows) or ~/.config/github-copilot/hosts.json (Linux/macOS)
2. Exchanges it for a short-lived API token via GitHub's internal token endpoint
3. Caches the API token with TTL-aware refresh (thread-safe, SemaphoreSlim double-check locking)
4. Injects auth headers into every SK request via a DelegatingHandler

Available Models

Copilot provides access to multiple AI providers through a single subscription:

Discover models at runtime:

var discovery = new CopilotModelDiscovery(provider, httpClient, logger);
var models = await discovery.DiscoverModelsAsync();

Configuration

Options

var kernel = Kernel.CreateBuilder()
    .UseCopilotChatCompletion(CopilotModels.Gpt4o, options =>
    {
        // Override token file location
        options.TokenFilePath = "/custom/path/apps.json";

        // GitHub Enterprise Server
        options.GitHubHost = "github.enterprise.com";

        // Enterprise SSL bypass
        options.DangerouslyDisableSslValidation = true;

        // Custom endpoint
        options.CustomEndpoint = "https://proxy.internal/copilot";
    })
    .Build();

Dependency Injection

services.AddCopilotAuthentication(configuration);
// or
services.AddCopilotAuthentication(o => o.OAuthToken = "ghu_...");

Environment Variables

Packages

Sample CLI: jdcplt

dotnet tool install -g JD.Tools.CopilotChat

# Interactive chat
jdcplt

# Single prompt
jdcplt --prompt "Explain async/await in C#"

# Choose a model
jdcplt --model claude-sonnet-4 --prompt "Write a haiku about code"

# List available models
jdcplt --list-models

# Enterprise SSL bypass
jdcplt --insecure

Documentation

Full documentation is available at GitHub Pages or in the docs/ directory:

• Getting Started
• Authentication Flow
• Configuration Reference
• Model Discovery

Related Projects

• JD.SemanticKernel.Connectors.ClaudeCode — Same pattern for Claude Code subscriptions

Both connectors implement the shared JD.SemanticKernel.Connectors.Abstractions interfaces, enabling MCP server bridging across subscriptions.

Building

git clone https://github.com/JerrettDavis/JD.SemanticKernel.Connectors.GitHubCopilot.git
cd JD.SemanticKernel.Connectors.GitHubCopilot
dotnet build
dotnet test

Contributing

See CONTRIBUTING.md for guidelines.

License

MIT