LLM Setup Troubleshooting

Troubleshoot MLJAR Studio connections to OpenAI, OpenRouter, Ollama, LM Studio, Jan, vLLM, llama.cpp, and custom OpenAI-compatible servers.

Provider cannot be saved

You need to click Test connection before saving a provider. MLJAR Studio tests whether the provider can be reached and whether the selected model is available. The Save provider button becomes active only after the test succeeds.

After you click Save provider, you should see a success toast. The active provider then appears in the top chip in the sidebar with a green dot.

Provider settings are locked

AI Provider Settings require a lifetime MLJAR Studio license. Activate your license if the provider selector is disabled. If access verification fails, use Retry in the settings panel.

Local server connection refused

Start the model server before refreshing models. In LM Studio, use the Developer tab; in Jan, open Settings → Local API Server. For vLLM and llama.cpp, keep the server process running and wait for model loading to finish.

Match the host and port to your running server. Loopback addresses point to the notebook backend machine. If the server runs on another machine or in a container, check that the endpoint is reachable from the backend.

Local server authentication fails

Use the model server’s key. The Jan preset requires a Jan server API key; configure a key in Jan and copy the same value into Studio. LM Studio, vLLM, llama.cpp, and Custom OpenAI-compatible allow an empty key only when the server does not require authentication.

Model list is empty or a model is missing

Check that a chat model is downloaded and exposed by the server. Click Refresh models and select a returned model identifier. OpenAI-compatible discovery requires a valid /models response relative to the base URL; a server that only implements chat generation cannot pass model discovery and testing.

404 or unsupported API format

For the default local presets, keep /v1 at the end of Base URL. Do not include /models or /chat/completions there. For Custom OpenAI-compatible, select Chat Completions unless your server supports the Responses endpoint. Match any custom API prefix configured on the server.

Connection test succeeds but generation fails

For OpenAI-compatible servers, Test connection validates model discovery and selection. Try a short notebook request to check generation. Inspect server logs for missing chat templates, unsupported streaming or tool calling, context limits, or insufficient memory. See Custom OpenAI-compatible Server for API requirements.

OpenRouter errors

• Check that the key belongs to OpenRouter and has not been revoked.
• If model discovery fails, verify network access and retry after entering a valid key.
• For model or usage errors, check your account limits, credits, and model availability. Select another available model, then test and save again.

Follow OpenRouter Integration for setup.

OpenAI API key error

If OpenAI authentication fails, check the following:

• The API key was copied completely.
• The API key was not revoked in the OpenAI dashboard.
• The OpenAI account has billing enabled if required.
• The selected model is available to your account.
• There are no extra spaces before or after the key.

OpenAI invalid model error

An invalid model error usually means the model name is wrong or your account does not have access to that model. For example, you can test with gpt-5.4 if it is available in your account. Try a model name you know is enabled, then click Test connection again.

Ollama connection refused

This usually means MLJAR Studio cannot reach the Ollama server. Check that Ollama is installed and running.

ollama --version
ollama list

Also check that the endpoint is correct. The default local endpoint is usually:

http://localhost:11434

Ollama model not found

If the model is missing, download it first. The model name in MLJAR Studio should match the model available in Ollama.

ollama pull qwen3.5:27b
ollama pull gemma4:31b
ollama list

Local model is too slow

Local inference speed depends on your hardware and model size. If the model is too slow, try a smaller model, close other heavy applications, or use a remote provider such as OpenAI or an Ollama-compatible cloud endpoint.

Ollama Cloud does not connect

• Check that the Ollama API key is correct.
• Check that the model name is correct, for example qwen3.5:397b or gemma4:31b.
• Confirm that the model is available in your Ollama Cloud account.
• Click Test connection again before saving the provider.
• Ask your IT or infrastructure team whether a proxy, firewall, or VPN rule is blocking access.

Setup guides for additional local servers: LM Studio, Jan, vLLM, and llama.cpp.

Still not working?

Review the provider setup pages again: OpenAI Integration, Ollama Local Setup, and Ollama Cloud Setup.

Frequently asked questions