OpenCode Qwen Auth Plugin

Qwen OAuth authentication plugin for OpenCode with multi-account rotation, proactive token refresh, and automatic API translation.

Features

• Device Flow OAuth - PKCE-secured authentication, works in headless/CI environments
• Multi-Account Support - Store and rotate between multiple Qwen accounts
• Proactive Token Refresh - Automatically refresh tokens before expiry
• Rate Limit Handling - Detects 429 responses, rotates accounts, respects retry-after
• API Translation - Bridges OpenAI Responses API ↔ Chat Completions API
• Streaming Support - Full SSE transformation for real-time responses

Installation

# Using Bun (recommended)
bun add opencode-qwen-auth

# Using npm
npm install opencode-qwen-auth

Add to your opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-qwen-auth"],
  "provider": {
    "qwen": {
      "npm": "@ai-sdk/openai",
      "options": {
        "baseURL": "https://portal.qwen.ai/v1",
        "compatibility": "strict"
      },
      "models": {
        "qwen3-coder-plus": { "contextWindow": 1048576 },
        "qwen3-vl-plus": { "contextWindow": 262144, "attachment": true }
      }
    }
  }
}

Quick Start

1. Start OpenCode in your project directory:

   opencode

2. Authenticate with Qwen:

   /auth

   Select Qwen OAuth and follow the device flow instructions.

3. Start coding with Qwen models:

   /model qwen/qwen3-coder-plus

Configuration

Create .opencode/qwen.json in your project or ~/.config/opencode/qwen.json for user-level config:

{
  "base_url": "https://portal.qwen.ai/v1",
  "client_id": "f0304373b74a44d2b584a3fb70ca9e56",
  "oauth_base_url": "https://chat.qwen.ai",
  "rotation_strategy": "round-robin",
  "proactive_refresh": true,
  "refresh_window_seconds": 300,
  "max_rate_limit_wait_seconds": 300,
  "quiet_mode": false
}

Configuration Options

Environment Variables

All options can be overridden via environment variables:

• QWEN_API_BASE_URL
• QWEN_OAUTH_CLIENT_ID
• QWEN_OAUTH_BASE_URL
• QWEN_ROTATION_STRATEGY
• QWEN_PROACTIVE_REFRESH
• QWEN_REFRESH_WINDOW_SECONDS
• QWEN_MAX_RATE_LIMIT_WAIT_SECONDS
• QWEN_QUIET_MODE

Models

Available via OAuth

Multi-Account Rotation

Add multiple accounts for higher throughput:

1. Run /auth and complete the first login
2. Run /auth again to add additional accounts
3. The plugin automatically rotates between accounts

Rotation Strategies

• round-robin: Cycles through accounts on each request
• sequential: Uses one account until rate limited, then switches

How It Works

This plugin bridges OpenCode's Responses API format with Qwen's Chat Completions API:

OpenCode → [Responses API] → Plugin → [Chat Completions] → Qwen
                                ↓
OpenCode ← [Responses API] ← Plugin ← [Chat Completions] ← Qwen

Request Transformation

Response Transformation (Streaming)

Converts SSE events from Chat Completions to Responses API format:

• response.created
• response.output_item.added
• response.content_part.added
• response.output_text.delta
• response.completed

Storage Locations

Security Note: Tokens are stored with restricted permissions (0600). Ensure appropriate filesystem security.

Troubleshooting

Authentication Issues

"invalid_grant" error

• Your refresh token has expired. Run /auth to re-authenticate.

Device code expired

• Complete the browser login within 5 minutes of starting /auth.

Rate Limiting

Frequent 429 errors

• Add more accounts with /auth
• Increase max_rate_limit_wait_seconds in config

Reset Plugin State

To start fresh, delete the accounts file:

rm ~/.config/opencode/qwen-auth-accounts.json

Development

This project uses Bun for development.

Prerequisites

• Bun 1.0+ (recommended)
• Node.js 20+ (for npm compatibility)

Getting Started

# Install dependencies
bun install

# Build
bun run build

# Run tests
bun test

# Run tests in watch mode
bun test --watch

# Run e2e test (requires authenticated Qwen account)
bun run test:e2e

# Link for local testing
bun link

Using npm

The project also works with npm:

npm install
npm run build
npm test

Known Limitations

• Audio input (input_audio) is not supported by Qwen and is converted to placeholder text

License

Apache-2.0