Qwen's web UI, reverse-engineered into an API buffet

What it does

qwen2API is a Go-based gateway that sits between your AI clients and Alibaba’s Qwen web interface. You feed it Qwen account credentials, and it exposes standard API surfaces — OpenAI Chat Completions, Anthropic Messages, Gemini GenerateContent — while handling the messy work of browser automation, account rotation, rate-limit cooling, and request retries behind the scenes. It also bundles a React WebUI for managing accounts, API keys, and testing chat or image generation without touching a terminal.

The interesting bit

The project doesn’t just proxy requests; it reverse-engineers Qwen’s web capabilities into protocol shapes that existing tooling already speaks. The model list endpoint dynamically fetches from upstream and synthesizes mode variants (-thinking, -deep-research, -image) as pseudo-models, letting clients toggle behaviors through model selection rather than custom parameters. Embeddings are acknowledged as fake — deterministic simulated vectors — which is refreshingly honest for a compatibility shim.

Key highlights

• Multi-protocol facade: OpenAI, Anthropic Claude, and Gemini APIs from one gateway
• Account pool with concurrency limits, jitter, and automatic retry on rate limits
• Image generation via POST /v1/images/generations with fixed size/ratio presets
• Built-in React admin console; backend hosts static frontend artifacts directly
• Multi-arch Docker images (amd64/arm64) published via GitHub Actions
• Health and readiness probes (/healthz, /readyz) for container orchestration

Caveats

• Embeddings are explicitly simulated placeholders, not real vectors
• OpenAI Assistants, Realtime, Audio, Batch, and Fine-tuning APIs are unimplemented
• File upload only covers the attachment pipeline; not a full OpenAI Files lifecycle
• Requires active Qwen web accounts and ongoing maintenance of browser automation

Verdict

Worth a look if you’re self-hosting AI infrastructure and need to unify Qwen access behind familiar API contracts. Skip it if you need native embeddings, voice features, or stateful OpenAI Assistants — or if maintaining scraper-adjacent tooling against a web UI sounds like a hobby you didn’t want.