[sosidudku1]

Ollama, LM Studio and Atomic Chat run on your own computer, but first-run setup listed them under Cloud models. People who already run one read the Cloud card as not theirs and the Local card as a big download (ATO-816). A Discord user also asked on day one: "can't I point it at a model I already have downloaded?" (ATO-1402).

What changes

• Cloud models lists cloud providers only. The "Local on this Mac" group is gone, and searching "ollama" there finds nothing.
• Local models gets two cards under the model list, also while the list is loading or nothing fits:
  • Use an app you already have opens a list of exactly Atomic Chat, Ollama and LM Studio. Back (or Esc) returns to the model list.
  • Use a .gguf file I already have opens a file dialog. The file is registered where it lies and no model is downloaded or copied (on a fresh install the engine itself is still fetched). The model starts on the built-in engine. The card works while the model list is still loading. It writes the same entry the TUI writes for a pasted path (desktop/main/local-gguf.ts is a copy of src/local-llm/catalog/local-gguf.ts, like huggingface.ts).
• The "Already run Ollama or LM Studio? Use it instead" hint on the first screen is removed.
• Card copy: Local "Private and free on this Mac. Download a model or use an app you have." Cloud "Connect a provider with your key or subscription."
• The finish screen says Model ready after an app, Cloud model ready after a key. The outcome stays cloud, so analytics does not change.

Checks added

• Wizard smoke (main.ts): no apps in the Cloud list or its search, one way in on Local, exactly three apps behind it, the screen title, Back lands on local_pick, no hint on the first screen.
• test/onboarding-mouse.mjs: the same path with real clicks; the Tab loop allows for the two new cards.
• test/onboarding-local-ready.test.mjs: "Model ready" for an app.
• test/onboarding-gguf-pick.test.mjs: the .gguf pick opens the dialog while the list loads, cancel, a refused file, no second dialog, a running download.

Not done here

• Settings › Models and Fusion labels are in a separate PR (ATO-1401).
• After "Local models" the flow records managed mode before the app is chosen, as it does today for "set up cloud meanwhile". Worth confirming on a real run that the chat goes to Ollama and the built-in engine does not start.

Linear: ATO-1309, ATO-1310, ATO-1311, ATO-1312, ATO-1313, ATO-1402

Not yet checked on the second Mac.