LLM Gateway Troubleshooting
Errors from the gateway come back as JSON in the same format OpenAI uses, so SDKs show a readable message. The code field tells you what went wrong:
| Status | Code | What to do |
|---|---|---|
| 401 | missing_api_key | Send your key as Authorization: Bearer YOUR_KEY. |
| 401 | invalid_api_key | The key is wrong, was replaced, or belongs to another gateway. Check you copied all of it, or issue a new one. |
| 401 | expired_api_key | Issue a new key. |
| 403 | gateway_suspended | The gateway is suspended. Contact support. |
| 429 | rate_limit_exceeded | Too many requests per minute. Wait for the time in Retry-After, then try again. |
| 523 | connector_unreachable | The connector or your model server isn't answering. See below. |
| 523 | gateway_not_placed | The connector has never connected. Install it with a fresh install command. |
The connector shows Offline, or you get 523
- Check the connector is running on your machine:
systemctl status cdnshark-tunnel # Linux cdnshark-tunnel status # macOS and Windows - Check the model server answers on that machine. For Ollama:
If that fails, start your model server. If the model runs on another machine, use its LAN address and make sure it listens on the network.curl http://localhost:11434/api/tags - Check the Model server box on the gateway page has the right host and port. See Change the Model Server.
- Make sure the network allows outbound UDP. Some strict office or guest networks block it.
The model list is empty in my tool
Your tool probably loads the list without sending the key. Type the model name by hand, or switch on Allow the model list without an API key. See Use Your Gateway with Cline or Open WebUI.
Answers come back as JSON or don't make sense
That is the model, not the gateway. Small models struggle with coding assistants' long instructions. Use a larger model and a bigger context window, as described in Use Your Gateway with Cline or Open WebUI.
Responses are slow
Speed depends on your machine's GPU or CPU and on your internet upload speed. Try a smaller or more compressed (quantized) model, or a faster connection.
"model not found"
The model name must match exactly what your model server calls it. With Ollama, run ollama list to see the names, including the tag, such as llama3.2:1b.