Qwen3 Call Patch Proxy

A robust HTTP proxy server that fixes malformed tool calls from Qwen3-Coder LLM models before sending them to OpenCode or other downstream services.

Note: This proxy is primarily developed for OpenCode, but may be useful for other LLM environments requiring tool call format fixes.

What It Does

Fixes Qwen3-Coder tool call issues:

• Consolidates fragmented tool calls across multiple SSE events
• Converts string parameters to proper types (arrays, booleans)
• Adds missing required parameters with sensible defaults
• Generates proper tool call IDs in OpenCode format

Demo

OpenCode successfully creating Conway's Game of Life using Qwen3-Coder through the proxy

Tested Models

• unsloth/Qwen3-Coder-30B-A3B-Instruct
• cpatonn/Qwen3-Coder-30B-A3B-Instruct-AWQ

Quick Start

Prerequisites

• Python 3.8+
• A running Qwen3-Coder model server (typically on port 8080)
• OpenCode or compatible client

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/qwen3-call-patch-proxy.git
   cd qwen3-call-patch-proxy

2. Install dependencies:

   pip install -r requirements.txt

3. Start the proxy:

   python call_patch_proxy.py

4. Configure OpenCode to use the proxy: Point OpenCode to http://localhost:7999 instead of your model server directly.

Architecture

OpenCode ──→ Proxy (7999) ──→ Qwen3 Server (8080)
             [Fixes tool calls]

OpenCode Configuration

To use this proxy with OpenCode, add the following provider configuration to your OpenCode settings:

{
  "provider": {
    "qwen3-coder-30b-a3b-instruct-vllm": {
      "name": "Qwen 3 Coder 30B (vLLM)",
      "npm": "@ai-sdk/openai-compatible",
      "models": {
        "cpatonn/Qwen3-Coder-30B-A3B-Instruct-AWQ": {}
      },
      "options": {
        "baseURL": "http://127.0.0.1:7999/v1"
      }
    }
  }
}

This configuration:

• Uses the proxy on port 7999 as the baseURL
• Works with OpenAI-compatible API format
• Supports AWQ quantized models for efficient inference

Health Monitoring

• Health check: GET http://localhost:7999/_health
• Reload config: POST http://localhost:7999/_reload

Documentation

For detailed information, see:

• Detailed Guide - Complete documentation with examples and configuration options

Contributing

1. Fork the repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Make your changes and add tests
4. Commit: git commit -m 'Add amazing feature'
5. Push: git push origin feature/amazing-feature
6. Open a Pull Request

License

This project is licensed under the GPL-3.0 License - see the LICENSE file for details.

Acknowledgments

• Built for the OpenCode ecosystem
• Designed to work with Qwen3-Coder models from Alibaba Cloud
• Inspired by the need for robust LLM tool call handling

---

Need help? Open an issue or check the discussions section!