Troubleshooting

Open System Settings → Apple Intelligence & Siri and turn Apple Intelligence on. Back in String Hopper's Settings, click Check again.

macOS downloads the model after you turn Apple Intelligence on. Leave your Mac on Wi-Fi and power for a while, then click Check again. Your closet and tags work in the meantime.

Apple Intelligence isn't supported on this Mac. You can still use everything by connecting a local model — including one running on another machine on your network.

Your closet is larger than Apple's model can read at once. Open the lens and narrow it under Runs on — by tag, by category, or with Choose pieces…. If this happens often, a local model with a larger context window can read more.

Apple's model declined something in the brief or the closet. Rephrase the lens in plainer terms and try again.

Apple's model only works in some languages. Write the brief in a supported language, or use a local model.

You chose A local model you run but one of the two is empty. Follow the five steps.

Use a full address such as http://localhost:1234/v1. The ? button beside the field lists the usual ones.

The server isn't running, or it's on a different port.

• Start the server — in LM Studio that's the Developer tab; for Ollama, open the app.
• Check the port matches: 1234 for LM Studio, 11434 for Ollama, 8080 for llama.cpp, 8000 for MLX and vLLM.
• Make sure the address ends in /v1.

The server is running but slow. A model that's still loading can cause this once — try again. If it keeps happening, the model is too large for your Mac; use a smaller one.

The server answered. Choose a model from the Model list in Settings. You also see this if the model you chose earlier is no longer loaded on the server.

The server refused the request. The text after the number is the server's own explanation.

• 401 or 403 — the server requires a key. Enter it under Key.
• 404 — wrong address, or the model name doesn't exist on the server.
• 400 when adding a photo — often the model can't see images. String Hopper tries again without the photo by itself; if you still see this, the server refused that too.
• 500 — often the server running out of memory. Try a smaller model.

The model stopped before writing an answer — commonly a thinking model that spent its whole budget reasoning, or a context window that's too small. Raise the context length on the server to 8,000 tokens or more and try again.

The model answered, but not in the structure the app asked for. Small models do this occasionally.

• Run the lens again. It often works the second time.
• Narrow the lens under Runs on so the model reads fewer pieces.
• Try the next model size up, or a server that enforces the answer format, such as LM Studio or llama.cpp.

Either the closet has fewer than two pieces, or the tags and categories under Runs on narrowed it to fewer than two. Click Reset there, or add more pieces.

The model chose pieces that are not on the list, and too few valid pieces remained to make an outfit. Run it again. If it keeps happening on a local model, the model is probably too small.

A capsule needs six to ten pieces that work together. With a small closet, or one narrowed too far, there may not be enough.

An edit lens returned no pieces. That can be correct for a small closet. Run it again to check.

The brief probably refers to data the app doesn't hold, or to a tag that isn't on your pieces. Check what a lens can read, and name fields and tags exactly.

The shape — a look, a capsule or an edit — is set when the lens is created. Delete the lens and create it again, saying the shape you want in the first few words.

Retake the photo with the piece laid flat on a plain surface that contrasts with it, in good light, one item per photo. You can always correct the colours by hand before saving.

Click Try again with a clearer photo. On a local model, Settings shows whether your model can look at photos — more here.

Your model or server isn't calling tools. Chat needs tool calling. Lenses do not. Choose a model labelled for tools, or switch the stylist to Apple's model for chat.

This is intended. Chat shows a card and does not save the piece. Type “add it” or “add both”, and chat saves it.

Check that the chat input bar shows Internet on. Some shops block automated reading or load details with scripts. In that case, describe the piece or add it from a photo instead.

If you've edited its instructions, click Instructions in the chat toolbar and Reset to default. New chat clears the conversation. It does not change your closet.