Works with opencode, crush, claude code router, roo code, cline and anything that speaks the OpenAI API. Has tool calling and streaming support.

⚠️ DEPRECATED — Qwen free usage has ended. You will get 401 errors. See GitHub Issue #3203. This project is no longer actively maintained.

New — Qwen CLI free quota is dead, but we have a great community on Discord that keeps up with AI news and free usage quotas. Join us.

New — Minimal terminal UI with full mouse support — lightweight, low resource usage.

New — TUI themes: dark, light, amber, contrast. Change from Settings or press t. Theme saves to ~/.local/share/qwen-proxy/config.json.

New — TUI selection style: solid or transparent. Transparent keeps row colors visible and uses a bold row + accent left marker.

New — coder-model now points to Qwen 3.6 Plus (Qwen team updated the alias). qwen3.5-plus requires Coding Plan subscription and won't work with OAuth accounts.

Discord

Important: Users may hit 504 / timeout errors at 130k–150k+ token contexts — this is a Qwen upstream limit.

For a serverless/edge alternative: qwen-worker-proxy

npm install -g qwen-proxy

qwen-proxy 

# and then add some accounts from the tui . full mouse Supported

Or headless (background/server mode):

qwen-proxy serve --headless

Point your client at http://localhost:8080/v1. API key can be any string.

git clone https://github.com/aptdnfapt/qwen-code-oai-proxy
cd qwen-code-oai-proxy
cp .env.example .env
docker compose up -d

The container mounts ~/.qwen from your host — accounts you add are picked up live by the running container without a restart.

Add an account while the container is running:

docker compose exec qwen-proxy node dist/src/cli/qwen-proxy.js auth add myaccount

Point your client at http://localhost:8080/v1.

npm install
npm run auth:add myaccount
qwen-proxy serve
# or headless:
npm run serve:headless

qwen-proxy serve                  # TUI dashboard
qwen-proxy serve --headless       # headless server

qwen-proxy auth list
qwen-proxy auth add <account-id>
qwen-proxy auth remove <account-id>
qwen-proxy auth counts
qwen-proxy usage

For fresh-machine regressions, use the built-in clean-home checks instead of ad-hoc shell probes:

npm run test:auth-clean-home
npm run test:first-run
npm run test:install-smoke

These scripts run the compiled code with a temporary HOME, so they simulate a new machine without touching your real ~/.qwen or local usage database.

More detail: docs/testing-clean-home.md

The easiest way to add accounts is from the TUI — just run qwen-proxy, go to the Accounts tab, and add from there.

You can also add via CLI:

qwen-proxy auth add account1
qwen-proxy auth add account2
qwen-proxy auth add account3

How rotation works:

• Requests rotate round-robin across all valid accounts
• Tokens refreshed ahead of expiry automatically
• Auth failures → one refresh attempt → rotate to next account
• Transient failures (429, 500, timeout) → rotate to next account, no cooldowns
• Client errors (bad payload etc.) → returned immediately, no rotation
• DEFAULT_ACCOUNT env var → that account is tried first
• Request counts reset daily at UTC midnight

For Docker:

docker compose exec qwen-proxy node dist/src/cli/qwen-proxy.js auth list
docker compose exec qwen-proxy node dist/src/cli/qwen-proxy.js auth add <account-id>
docker compose exec qwen-proxy node dist/src/cli/qwen-proxy.js auth remove <account-id>

Note: coder-model is an alias maintained by Qwen. It was Qwen 3.5 Plus, now updated to Qwen 3.6 Plus. Both qwen3.5-plus and qwen3.6-plus resolve to coder-model.

• POST /v1/chat/completions — Chat completions (streaming + non-streaming)
• GET /v1/models — List available models
• POST /v1/web/search — Web search (2000 req/day free)
• GET/POST /mcp — MCP server (SSE transport)
• GET /health — Health check

import OpenAI from 'openai';

const openai = new OpenAI({
  apiKey: 'fake-key',
  baseURL: 'http://localhost:8080/v1'
});

const response = await openai.chat.completions.create({
  model: 'coder-model',
  messages: [{ role: 'user', content: 'Hello!' }]
});

console.log(response.choices[0].message.content);

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer fake-key" \
  -d '{
    "model": "coder-model",
    "messages": [{"role": "user", "content": "Hello!"}],
    "temperature": 0.7,
    "max_tokens": 200,
    "reasoning": {"effort": "high"}
  }'

effort can be "high", "medium", "low", or "none" (disables thinking).

curl -X POST http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer fake-key" \
  -d '{
    "model": "coder-model",
    "messages": [{"role": "user", "content": "Explain how to reverse a string in JavaScript."}],
    "stream": true,
    "max_tokens": 300,
    "reasoning": {"effort": "medium"}
  }'

Free web search — 1000 requests/account/day:

curl -X POST http://localhost:8080/v1/web/search \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer fake-key" \
  -d '{"query": "latest AI developments", "page": 1, "rows": 5}'

Add to ~/.config/opencode/opencode.json:

effort can be "high", "medium", "low", or "none" (disables thinking). change it from the ctrl-t key bind on opencode

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "qwen": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "proxy",
      "options": {
        "baseURL": "http://localhost:8080/v1"
      },
      "models": {
        "coder-model": {
          "name": "qwen3.6-plus" ,
           "reasoning": true,
          "modalities": {
          "input": [
            "text",
            "image"
          ],
          "output": [
            "text"
          ]
        },
            "attachment": true,
        "limit": {
          "context": 195000,
          "output": 60000
          }

        }
      }
    }
  }
}

Add to ~/.config/crush/crush.json:

{
  "$schema": "https://charm.land/crush.json",
  "providers": {
    "proxy": {
      "type": "openai",
      "base_url": "http://localhost:8080/v1",
      "api_key": "",
      "models": [
        {
          "id": "coder-model",
          "name": "coder-model",
          "cost_per_1m_in": 0.0,
          "cost_per_1m_out": 0.0,
          "cost_per_1m_in_cached": 0,
          "cost_per_1m_out_cached": 0,
          "context_window": 150000,
          "default_max_tokens": 32768
        }
      ]
    }
  }
}

{
  "LOG": false,
  "Providers": [
    {
      "name": "qwen-code",
      "api_base_url": "http://localhost:8080/v1/chat/completions/",
      "api_key": "any-string",
      "models": ["coder-model"],
      "transformer": {
        "use": [
          ["maxtoken", {"max_tokens": 32768}],
          "enhancetool",
          "cleancache"
        ]
      }
    }
  ],
  "Router": {
    "default": "qwen-code,coder-model"
  }
}

1. Go to settings → choose OpenAI Compatible
2. Set URL: http://localhost:8080/v1
3. API key: any random string
4. Model: coder-model
5. Disable streaming checkbox (Roo Code / Kilo Code)
6. Max output: 32000
7. Context window: up to 300k (but past 150k gets slower)

Add to ~/.config/opencode/config.json:

{
  "$schema": "https://opencode.ai/config.json",
  "mcp": {
    "qwen-web-search": {
      "type": "remote",
      "url": "http://localhost:8080/mcp",
      "headers": {
        "Authorization": "Bearer your-api-key"
      }
    }
  }
}

Omit headers if you have no API key set. Works with other MCP clients too.

# Single key
API_KEY=your-secret-key

# Multiple keys
API_KEY=key1,key2,key3

Supported headers:

• Authorization: Bearer your-secret-key
• X-API-Key: your-secret-key

If no API key is configured, no auth is required.

Set via environment variables or .env file:

Compatibility aliases: DEBUG_LOG=true → LOG_LEVEL=debug, LOG_FILE_LIMIT → MAX_DEBUG_LOGS

Example .env:

LOG_LEVEL=debug
MAX_DEBUG_LOGS=10
API_KEY=your-secret-key
DEFAULT_ACCOUNT=my-primary-account

Port and host can also be changed from the TUI Settings screen and are saved to config.json automatically.

curl http://localhost:8080/health

Returns server status, account validation, token expiry info, request counts.

qwen-proxy usage
# or
npm run usage
npm run tokens

Shows daily token usage, cache hits, request counts per account. Also visible in the TUI Usage screen.

Change live without restart:

# inspect
GET /runtime/log-level

# change
POST /runtime/log-level
{"level": "debug"}

# change without persisting
POST /runtime/log-level
{"level": "error", "persist": false}