# LobeHub

LobeHub can add custom AI providers. Choose the OpenAI request format with the address up to /v1 and every NezhaGate model is available; self-hosted instances can use environment variables instead.

> https://nezhagate.com/en/docs/integrations/lobehub

Written against LobeHub 2.2

## Add a custom provider

1. Open **App Settings** → **AI Providers** and click **Add Custom Provider**.

2. Give a lowercase **Provider ID** (lowercase letters, digits, `-` and `_`; it cannot be changed later), choose **OpenAI** as **Request Format**, enter `https://nezhagate.com/v1` as **Proxy URL** and your key as **API Key**, then click **Create**.

3. On the provider page click **Fetch models**, or add models by hand with **Add Model**, setting the maximum context window and ticking tool calling and vision where the model has them.

4. Enable the models you want and pick one in a chat.

Proxy URL`https://nezhagate.com/v1`

Keep **Use Client Request Mode** off (the NezhaGate API does not accept cross-origin browser requests), and keep **Use Responses API Specification** off as well.

## Claude (optional)

Add a second custom provider with **Anthropic** as **Request Format** and `https://nezhagate.com/anthropic` as **Proxy URL**.

## Self-hosted: environment variables

A self-hosted LobeHub can be pointed at NezhaGate with environment variables:

```
OPENAI_API_KEY=YOUR_API_KEY
OPENAI_PROXY_URL=https://nezhagate.com/v1
OPENAI_MODEL_LIST=-all,+gpt-5.5,+claude-sonnet-5
ANTHROPIC_API_KEY=YOUR_API_KEY
ANTHROPIC_PROXY_URL=https://nezhagate.com/anthropic
```

In `OPENAI_MODEL_LIST`, `-all` hides every built-in model, `+model-id` adds a NezhaGate model, and `+model-id=Display name` also sets its display name.

## Test

In **Connectivity Check** on the provider page, pick a model and click **Check** (it sends one short message); Check passed means you are done.

## Troubleshooting

**No reply, or an empty reply**

The proxy URL must include `/v1`: `https://nezhagate.com/v1`.

**Failed to fetch models**

Check the key and the proxy URL; you can also add models by hand with **Add Model**.

**CORS errors in the browser console**

Switch **Use Client Request Mode** off so the requests go out from LobeHub's server.