Add a provider
New to AI in Watchflows? Set Up AI walks you from a first launch to a working Flow Builder, one step at a time.
Settings ▸ AI Providers shows one card per configured provider, under a hero naming the Default Provider that every AI node uses unless it picks another. Add Provider opens the editor: choose a Type, name it, paste an API Key if the type needs one, then set a Default Model and a Default Image Model.
Watchflows does no AI billing. You bring the model; the provider bills you.
The ten types
| Type | API key | Where it runs |
|---|---|---|
| Ollama | Not needed | Your Mac, http://localhost:11434 |
| LM Studio | Optional | Your Mac, http://localhost:1234 |
| Apple Intelligence (on-device) | Not needed | Your Mac, nothing to install |
| Claude (Subscription) | Not needed | The claude CLI you installed |
| ChatGPT (Subscription) | Not needed | The codex CLI you installed |
| OpenAI | Required | api.openai.com |
| Anthropic (Claude) | Required | api.anthropic.com |
| Google (Gemini) | Required | generativelanguage.googleapis.com |
| OpenRouter | Required | See below |
| OpenAI Compatible | Required | Any OpenAI-shaped endpoint you name |
Only Ollama, LM Studio, and OpenAI Compatible have an editable Base URL. The rest are fixed.
Text and image defaults
A node left on its provider's default routes by what the payload holds: text-only to the Default Model, image-carrying to the Default Image Model. Pick a model on the node itself and that choice always wins.
OpenRouter routing
OpenRouter adds a Routing picker — Global, EU, or US. EU and US keep inference on providers inside that region and need an OpenRouter Business plan. There is no free-text base URL for this provider; only the three OpenRouter hosts are ever used, and anything else stored against the card reads back as Global.
| Routing | Host |
|---|---|
| Global | openrouter.ai |
| EU | eu.openrouter.ai |
| US | us.openrouter.ai |
Good to know
Keys live in your login keychain. They are stored under the com.watchflows.app service, never written into a flow file, and never included in an export.
Apple Intelligence has requirements. It needs macOS 26 or later on an Apple-silicon Mac with Apple Intelligence turned on in System Settings. Until then the type stays in the picker but reports why it cannot run.
Start Ollama or LM Studio first. A server that is not up fails the node with a connection error. A token-protected LM Studio takes its token in the optional API Key field.