Plug one OpenAI-compatible endpoint into LiteLLM
LiteLLM speaks OpenAI by default. Register this gateway as an OpenAI-compatible provider — one base URL, one key — and route GPT, Claude, Gemini, Kimi, GLM and DeepSeek through your existing LiteLLM setup.
This gateway as a LiteLLM provider vs. wiring each provider
LiteLLM can talk to many providers, but each needs its own credentials and config. Pointing it at one compatible gateway collapses that to a single entry.
| This gateway in LiteLLM | Per-provider in LiteLLM | |
|---|---|---|
| Provider entries to configure | One OpenAI-compatible entry | One block per provider with separate keys |
| Credentials to rotate | A single key | A key per upstream provider |
| Model routing | Switch by model string under one base URL | Map each model to its provider config |
| Chinese models | Kimi, Qwen, GLM, DeepSeek on the same entry | Separate setup, some not natively supported |
| Billing | Prepaid CNY credits, no expiry | Each provider billed separately |
| Provider-native quirks | Common OpenAI surface | Full per-provider parameter support |
Who this is for
A good fit if you
- Already run LiteLLM (library or proxy) and want fewer provider configs
- Want one key and base URL to cover frontier and Chinese models
- Prefer rotating a single credential over many provider keys
- Want prepaid credits instead of several metered bills
Not the right fit if you
- Need LiteLLM to pass provider-native parameters this gateway doesn't expose
- Must keep direct per-provider billing for cost attribution
- Require USD invoicing for procurement
- Depend on a model the gateway does not route
Honest limitations
What to expect when LiteLLM points at a compatible gateway.
- LiteLLM's provider-specific features only work if the underlying surface exposes them; a compatible endpoint covers the common OpenAI parameters.
- Billing and prepaid credits are in CNY; LiteLLM cost tracking will reflect gateway pricing, not upstream vendor pricing.
- Rate limits are per key — configure LiteLLM retries and backoff for HTTP 429.
- We do not publish latency numbers we have not measured. Benchmark through your LiteLLM setup before committing.
What you register in LiteLLM
The concrete values for an OpenAI-compatible provider entry. Multipliers apply to a 0.4 CNY per 1M token base.
Throughput and latency through LiteLLM depend on your deployment and region. We do not publish numbers we have not measured — run your own LiteLLM benchmark against the docs endpoint.
LiteLLM-compatible API — FAQ
How do I add this gateway to LiteLLM?
Register an OpenAI-compatible provider with api_base set to https://api.weeanno.example/v1 and your key as the api_key. Then reference models by their gateway model string.
Does it work with both the LiteLLM library and the proxy?
Yes. Anything that accepts an OpenAI-compatible base URL and key works — the Python library, the proxy server, and frameworks that wrap LiteLLM.
Can LiteLLM route Chinese models through it?
Yes. Because they share one OpenAI-compatible base URL, LiteLLM reaches Kimi, Qwen, GLM and DeepSeek by model string without separate provider setup.
How is cost tracked?
LiteLLM tracks usage against the gateway's per-model rates (a fixed multiplier on a 0.4 CNY per 1M token base), paid from prepaid credits that never expire.
One LiteLLM provider, every model
Buy a key, add one OpenAI-compatible entry, and route everything through LiteLLM.