Poly Model Bridge for Copilot

Bring any third-party model endpoint into GitHub Copilot Chat's model picker — over three wire protocols:

Mix protocols freely — even per model within one provider. Works in Chat, Edits and Agent mode, with free model switching from the picker.

Highlights

• 🎚️ Reasoning effort variants, Claude Code style — configure "efforts": ["low", "medium", "high"] on a model and each effort shows up as its own picker entry, e.g. GPT-5.5 (high). Mapped per protocol: output_config.effort (Anthropic) / reasoning_effort (Chat Completions) / reasoning.effort (Responses).
• 🧠 Thinking in the UI — set "thinking": true to stream Anthropic thinking deltas or Responses reasoning summaries into the chat.
• 🛠️ Tool calling on all three protocols (Copilot Agent mode ready)
• 🖼️ Image input on all three protocols
• 🔑 API keys per provider in VS Code SecretStorage (never in settings.json); x-api-key / Bearer styles; custom headers; key-less local endpoints supported
• 🧩 Multiple providers side by side, grouped names in the picker

Quick Start (no JSON required)

1. Command Palette → Poly Model Bridge: Add Provider — a wizard asks for a name, base URL, protocol and API key, then fetches the endpoint's model list so you just tick the models you want.
2. Copilot Chat model picker → Manage Models → tick the models under Poly Model Bridge.
3. Ongoing management (add/remove models, effort levels, thinking, protocol overrides, connection test): Poly Model Bridge: Manage Providers, also reachable from the provider's gear in Manage Models.

Advanced: configure via settings.json

The UI and JSON edit the same setting (polyBridge.providers) — power users can edit directly:

{
  "polyBridge.providers": [
    {
      "name": "MyRelay",
      "baseUrl": "https://my-relay.example.com",
      "apiType": "anthropic",                    // provider default protocol
      "models": [
        { "id": "claude-opus-4-8", "name": "Claude Opus 4.8", "efforts": ["high", "xhigh"] },
        { "id": "claude-sonnet-4-6", "name": "Claude Sonnet 4.6", "thinking": true },
        // per-model protocol override:
        { "id": "gpt-5.5", "name": "GPT-5.5", "apiType": "chat-completions", "efforts": ["low", "medium", "high"] }
      ]
    }
  ]
}

2. Command Palette → Poly Model Bridge: Set API Key → pick the provider → paste your key.
3. Copilot Chat model picker → Manage Models → tick the models under Poly Model Bridge.

Model options

Provider options: apiType, authHeader (x-api-key | authorization), requiresApiKey: false for local servers, extraHeaders.

FAQ

Models don't show up? Check polyBridge.providers is set, run "Chat: Manage Language Models", reload the window if needed.

401/403? Try the other authHeader style — official Anthropic uses x-api-key, most OpenAI-style gateways use Authorization: Bearer.

protocol_mismatch / provider errors on one model? That model isn't served over the configured protocol by your gateway — set the correct per-model apiType.

Inline completions? VS Code limitation: BYOK models power Chat/Edits/Agent only; ghost-text completions keep using Copilot's built-in model.

中文说明

把任意第三方模型端点接入 GitHub Copilot 模型选择器，支持三种协议：Anthropic Messages、OpenAI Chat Completions、OpenAI Responses，可按模型混用；支持 Chat / Edits / Agent 模式。

亮点：像 Claude Code 一样按模型配置多档 reasoning effort（"efforts": ["low","medium","high"]），每档在选择器里是独立条目（如 GPT-5.5 (high)）；"thinking": true 可把思考过程流式显示在聊天里；API Key 按 provider 存系统钥匙串。

快速上手：设置里配 polyBridge.providers（见上方英文示例）→ 命令面板 Poly Model Bridge: Set API Key → 模型选择器 Manage Models 勾选。

常见问题：401/403 换 authHeader 风格；某模型报 protocol_mismatch 说明网关没按该协议提供它，给这个模型单独设 apiType；行内补全不走 BYOK（VS Code 平台限制）。

License

MIT