Custom LLM Provider

Connect any OpenAI-compatible AI endpoint to GitHub Copilot Chat in Visual Studio Code.
Works out of the box with Alibaba DashScope (Qwen), OpenRouter, and any other OpenAI-compatible API.

🎯 Primary Use Case — Alibaba Cloud Coding Plan

This extension was developed primarily to bring Alibaba Cloud Coding Plan into Visual Studio Code.

Alibaba's Coding Plan feature in Model Studio lets you run powerful Qwen Coder models in full agent mode — editing files, running tests, searching your codebase — all from within GitHub Copilot Chat. This extension bridges the gap by exposing those models directly in the VS Code model picker.

✨ Features

• Models appear directly in the Copilot Chat model picker — no extra setup
• @qwen chat participant — invoke your custom models anywhere in Copilot Chat
• Multi-provider support — connect Alibaba DashScope, OpenRouter, and any other provider simultaneously, each with its own URL and API key
• Dynamic model discovery — models are fetched automatically from each provider's /v1/models endpoint on startup
• Full streaming support (Server-Sent Events)
• Tool calling support — agent mode, /fix, /edit, @workspace all work
• Automatic retry with exponential backoff for network failures and rate limits
• Hot-reload on settings change — no restart needed
• Zero runtime dependencies

🚀 Quick Start

1. Add your first provider

Open the Command Palette (Ctrl+Shift+P) and run Custom LLM: Add provider.

The wizard will ask for:

1. Provider name — e.g. Alibaba DashScope
2. Base URL — e.g. https://coding-intl.dashscope.aliyuncs.com/v1
3. API key — your sk-… key

After saving, the extension automatically fetches available models from the provider.

Get your API key from Alibaba Cloud Model Studio → API Keys section.
Note that Coding Plan API keys are separate from regular DashScope keys.

2. Pick a model in Copilot Chat

Open Copilot Chat (Ctrl+Alt+I) → click the model name → your models appear under Custom LLM.

First time only: Open Ctrl+Shift+P → Chat: Manage Language Models → hover over each model → click the eye icon 👁 to enable it in the picker.

3. Use the @qwen participant (optional)

Type @qwen in Copilot Chat to always route to your custom model, regardless of picker selection.

@qwen explain the auth flow in this codebase
@qwen qwen3-coder-plus refactor this function

🌐 Supported Endpoints

🔌 Multi-Provider Support

You can connect multiple providers at once — for example, use Alibaba DashScope and OpenRouter side by side. Each provider has its own URL and API key; models from all providers are merged into a single list in the Copilot Chat picker.

Add a provider:

Ctrl+Shift+P → Custom LLM: Add provider

Manage providers (edit / remove):

Ctrl+Shift+P → Custom LLM: Manage providers

Refresh the model list:

Ctrl+Shift+P → Custom LLM: Refresh model list from API

⚙️ Settings

Provider configuration

Providers are stored in the customLlm.providers array. Each entry has three fields:

You can also edit settings directly in User Settings JSON (Ctrl+Shift+P → Open User Settings (JSON)):

"customLlm.providers": [
  {
    "name": "Alibaba DashScope",
    "baseUrl": "https://coding-intl.dashscope.aliyuncs.com/v1",
    "apiKey": "sk-YOUR-KEY-HERE"
  },
  {
    "name": "OpenRouter",
    "baseUrl": "https://openrouter.ai/api/v1",
    "apiKey": "sk-or-YOUR-KEY-HERE"
  }
]

Legacy settings (customLlm.baseUrl and customLlm.apiKey) are automatically migrated to the new customLlm.providers format on the first startup after updating to v0.4.0. No manual action needed.

Model list

customLlm.models is auto-populated by model discovery and does not normally need to be edited manually. The extension merges discovered models with any existing entries — custom entries are preserved.

Default models

If no providers are configured or the API is unreachable, the extension falls back to these built-in defaults (all available in Alibaba Cloud Coding Plan):

🛠️ Commands

📋 Requirements

• Visual Studio Code 1.104.0 or later
• GitHub Copilot extension installed and signed in (individual plan)
• An API key for your chosen provider

🔁 Retry Behavior

The extension automatically retries failed requests with exponential backoff:

Maximum delay capped at 10 seconds. Request cancellation is never retried.

⚠️ Known Limitations

• GitHub Copilot Coding Plan (GitHub's native multi-step agent mode) is tied to GitHub's own infrastructure and cannot use custom providers. Use Alibaba Cloud Coding Plan as a powerful alternative.
• Inline completions (ghost text) are provided by GitHub Copilot and cannot be redirected
• Models only appear in the picker on individual GitHub Copilot plans (not Business/Enterprise)

🛠️ Troubleshooting

Agent mode loops / repeating the same actions

If the model keeps calling the same tool in a loop, switch to a larger variant — smaller models sometimes struggle with complex multi-step tool orchestration. Try qwen3-coder-plus or qwen3-max instead of lighter models.

Models don't appear in the picker

1. Open Ctrl+Shift+P → Chat: Manage Language Models
2. Hover over each model in the Custom LLM section
3. Click the eye icon 👁 to make it visible in the picker

If the section doesn't appear at all, check that the extension is active: Ctrl+Shift+P → Extensions: Show Installed Extensions and verify Custom LLM Provider is enabled.

To force a model refresh: Ctrl+Shift+P → Custom LLM: Refresh model list from API.

401 Unauthorized — "invalid access token or token expired"

Your API key is missing or incorrect. Fix:

1. Open Ctrl+Shift+P → Custom LLM: Manage providers
2. Select your provider → Edit
3. Paste your API key (starts with sk-)
4. Make sure there are no extra spaces around the key

Get your key from Alibaba Cloud Model Studio → API Keys section.

Requests fail with 404 or empty responses

Check that the baseUrl ends with /v1 and the model id values match exactly what your provider expects. For DashScope, use https://coding-intl.dashscope.aliyuncs.com/v1.

Settings changes not taking effect

The extension hot-reloads on settings change, but it may take a few seconds. If models still don't update, reload the VS Code window: Ctrl+Shift+P → Developer: Reload Window.

Migrating from v0.3.x or earlier

The old customLlm.baseUrl and customLlm.apiKey settings are automatically migrated to the new customLlm.providers array on the first launch. If you need to re-run migration manually, remove the customLlm.providers entry from your settings and reload VS Code.

🔨 Building from Source

# Clone the repository
git clone https://github.com/milhaus123/vscode-custom-llm-provider.git
cd vscode-custom-llm-provider

# Install dev dependencies
npm install

# Compile TypeScript
npm run compile

# Package as .vsix
npx vsce package

# Install locally
code --install-extension vscode-custom-llm-provider-*.vsix

💖 Support the Project

If this extension saves you time, consider buying me a coffee!

Your support helps keep the project maintained and updated with new model releases. 🙏

📄 License

MIT © 2026 Martin Říha