CLLMs for Copilot Chat
======================

This project is a derivative work, adapted into a multi-provider extension for
Chinese frontier LLMs (Qwen, z.ai GLM, MiniMax) and other OpenAI-compatible
providers, of:

  deepseek-v4-for-copilot
  Copyright (c) 2026 Vizards
  https://github.com/Vizards/deepseek-v4-for-copilot
  Licensed under the MIT License.

The original project established the architecture used here: registering a
BYOK model into the GitHub Copilot Chat model picker through the native
vscode.LanguageModelChatProvider API, including the streaming request
pipeline, the vision proxy bridge, thinking-mode handling, prompt-cache
diagnostics, and internationalization.

Modifications in this project include (non-exhaustive):
  - A multi-provider architecture with first-class Qwen (DashScope), z.ai
    (Zhipu GLM), and MiniMax providers, each with its own base URL, API key,
    and model IDs; fully configurable for other compatible providers.
  - Per-provider thinking control: Qwen's `enable_thinking` / `thinking_budget`,
    GLM's `thinking: { type }`, and MiniMax's `thinking: { type: "adaptive" }`
    with `reasoning_split`.
  - Hybrid vision: native image input (Qwen-VL, GLM-4.5V) plus the Copilot-model
    proxy fallback for text-only models.
  - DashScope cache-usage accounting (prompt_tokens_details.cached_tokens).
  - Rebranding to CLLMs, multi-provider model registry, pricing, and documentation.

Grateful thanks to the original author. The MIT License text covering both the
original work and this adaptation is in the LICENSE file.
