LM Custom Provider

Add any OpenAI / Anthropic / Gemini / Azure OpenAI compatible provider straight into the GitHub Copilot Chat model picker. Bring your own keys, your own models, your own providers — all behind a single, native-looking management panel.

Features

• Multi-provider — five protocol drivers built in:
  • OpenAI Compatible (DeepSeek, OpenRouter, Groq, Together, SiliconFlow, Moonshot, …)
  • Anthropic Compatible (Claude native /v1/messages)
  • Google Gemini (generativelanguage.googleapis.com)
  • Azure OpenAI (deployment-based routing)
  • AWS Bedrock (placeholder, coming soon)
• Native UI — a Lit + @vscode-elements/elements webview that matches the VS Code theme.
• Two ways to add models — fetch the live /models list and click +, or enter the metadata manually for providers without a discovery endpoint.
• Flexible presets — choose from built-in presets (DeepSeek, Anthropic, Gemini, etc.) or use the blank custom preset to configure any provider manually. Type and base URL are optional for custom configurations.
• Per-model capability flags — tools, vision, and thinking toggles surface the matching VS Code Chat capabilities (including the per-model thinking-effort dropdown).
• Hot reload — editing a provider takes effect on the next chat turn; in-flight streaming responses keep their original config snapshot and finish naturally.
• Settings Sync — provider definitions (without API keys) are stored in settings.json and automatically synced across devices via VS Code Settings Sync.
• Per-provider key storage — choose SecretStorage (default, never leaves the device) or Settings (sync-able but visible to everyone with read access to your config).
• Anthropic auth modes — support for different authentication methods for Anthropic-compatible providers.
• Reasoning cache — for thinking-capable models, the same multi-turn reasoning_content re-injection logic generalised across providers.

Quick start

1. Install the extension.
2. Run LM Custom Provider: Open Provider Manager from the command palette.
3. Click +, pick a preset (e.g. DeepSeek or Anthropic) or use Custom for manual configuration, enter your API key, save.
4. Click Fetch From Remote (OpenAI-compatible providers) to import models — or click Add Manually for everything else.
5. Open Copilot Chat → model picker → your provider's models appear under their own vendor.

Configuration

Develop

# Required: Node 24 (see .node-version) and pnpm via corepack.
corepack enable
pnpm install
pnpm run compile

Press F5 to launch the Extension Development Host.

Architecture overview

extension.ts
 ├─ ConfigStore (settings.json + SecretStorage + VS Code Settings Sync)
 ├─ ProviderRegistry → AggregateChatProvider → vscode.lm.registerLanguageModelChatProvider (vendor: ccp)
 │   ├─ per-provider model ids: <providerId>::<modelId>
 │   └─ ProtocolDriver (OpenAI / Anthropic / Gemini / Azure / Bedrock)
 └─ ProviderManagerPanel (Lit webview)

A single static vendor copilot-custom-provider is registered. All configured providers are exposed as models under that vendor with composite ids <providerId>::<modelId>. In-flight streams capture a RequestSnapshot and complete with their original driver + config + key, so saving config never interrupts the user.

License

MIT