Start with the visible symptom, isolate the layer that failed, and change one thing at a time. Find the steps for memory errors, slow responses, app connections, and PDF chat below.
Loading troubleshooting steps…
Start with the symptom
Copy the exact error, identify the failing layer, and test the smallest working path before changing multiple settings at once.
“Ollama is not recognized,” command not found: ollama, or a missing command in an editor terminal means that shell cannot find the executable. It does not establish that the model download, GPU, or Ollama server is broken. Check installation and the terminal's search path first.
Check first: Ollama is not installed or installation did not finish.
A downloaded model and a loadable model are different things. If LM Studio cannot load a file, first check the download type, available memory, and selected inference engine. A larger download or repeated reinstall does not fix an incompatible model architecture or an allocation failure.
Check first: The model format and selected runtime do not match.
If a local model loads but invents facts, ignores instructions, repeats itself, or gives an incomplete answer, separate missing information from model capability and configuration. A larger context setting does not supply knowledge the model never received, and a fluent answer is not evidence that it is correct.
Check first: The model is unsuitable for the task.
If Ollama reports “model not found,” stops while loading, or closes before answering, first separate a missing download from a stopped service and a memory failure. The error text decides which check comes next.
Check first: The model is missing or incomplete.
“Slow” can mean at least five different failures: the model takes too long to load, the first token is delayed, generation speed is low, performance collapses during a long chat, or the machine slows after sustained load. Diagnose which one you have before changing random settings.
Check first: The model is too large for the fast memory pool.
An out-of-memory failure usually means the complete workload exceeds the usable memory pool, not merely that the model file is large. Count model weights, runtime overhead, context, multimodal components, document processing, and every other application that shares the same RAM, dedicated GPU memory, or Apple unified memory.
Check first: The model weights leave no practical headroom.
Open WebUI is the interface. Ollama is the model runtime underneath it. Get Ollama working by itself first, then connect Open WebUI to the Ollama service. If models do not appear, diagnose the service address and network boundary instead of reinstalling both products at random.
Check first: Ollama is not running or does not have a working model.
A local PDF workflow has four separate jobs: extract the text, split and index it, retrieve the right passages, and let the model answer from those passages. When PDF chat fails, identify which job failed before changing the model.
Check first: Text extraction failed or produced a bad reading order.
A model running slowly does not establish that the GPU is missing. Check whether the operating system sees the graphics card, whether the runtime supports it, and whether this particular loaded model uses it.
Check first: The driver or backend is missing.
“Model not found,” a download stalled halfway, and a completed file that will not load are different failures. Keep the exact error and download source, then follow the matching check below.
Check first: The identifier or tag is wrong.
An app installed on your computer can still send prompts or documents to a remote model, embedding service, web tool, or synced folder. Check the exact workflow with a harmless sample before entering sensitive material.
Check first: Only the model runner was checked.