Set Up AI Assist
Dieser Inhalt ist noch nicht in deiner Sprache verfügbar.
AI Assist lets a language model draft cards, rewrite text, plan changes to your boards and answer questions about your story. The model only proposes. You review each proposal in FlowBoard, and nothing in your document changes until you apply it. A proposal you apply is a single step that Cmd+Z undoes. How AI Assist works explains what each mode can change and what a request sends to the model.
AI Assist is not available on iPad and iPhone. If Settings has no AI Assist section, your edition of FlowBoard doesn’t include it.
Choose where the model runs
Section titled “Choose where the model runs”FlowBoard doesn’t come with a model. You connect it to one:
- Local: a model server on your Mac, such as Ollama or LM Studio. You need no account and no key, and requests cost nothing. Your text stays on your Mac as long as the server runs on it.
- Cloud (bring your own key): a provider such as OpenRouter or Anthropic. You need an account and an API key from the provider, and the provider charges for each request. It receives the context that each request includes.
| Provider | Kind | Support |
|---|---|---|
| Ollama | Local | Tested |
| LM Studio | Local | Tested |
| Custom OpenAI-compatible (local) | Local | Compatible |
| OpenRouter | Cloud | Tested |
| Anthropic | Cloud | Tested |
| OpenAI-compatible provider | Cloud | Compatible |
Compatible means that FlowBoard speaks the provider’s protocol but the service itself has not been verified. OpenRouter offers many models through one key, including batch models that run requests in the background.
Add a connection
Section titled “Add a connection”- Open Settings (Cmd+,) and select AI Assist.
- Select Add connection.
- Choose a Provider. For Ollama, LM Studio, OpenRouter and Anthropic, FlowBoard fills in the Base URL. For the other two, enter the server’s address.
- Enter a Connection name, for example “Studio Mac” or “OpenRouter”.
- For a cloud provider, enter your API key.
- Select Test connection. If FlowBoard reaches the server, it says how many models the server offers, for example Reachable — 12 models.
- Under Models, type a model ID or choose one of the suggestions, then select Add. Add every model you want to switch between.
- Select Save connection.
Use the model’s API identifier, not its display name, for example qwen/qwen3.8-27b. For a
local server, download or load the model there first. If the test reports no models, the
server is running but has none available.
A connection is saved on this Mac, not in the document. With several connections, Make default chooses the one AI Assist starts with. Edit and Remove change or delete a connection.
Your API key
Section titled “Your API key”FlowBoard stores API keys in your Mac’s Keychain, never in the document or in its own settings. While editing a connection, Remove key deletes the stored key. Test connection uses a key you’ve typed without saving it.
The first request after you start FlowBoard reads the key. If macOS asks whether FlowBoard may use it, choose Always Allow so it doesn’t ask again.
Switch models
Section titled “Switch models”Each connection holds a list of models, and one of them is current. To change it:
- In the AI Assist panel, choose another Model next to the Connection. FlowBoard remembers the choice for your next new request.
- In Settings → AI Assist, edit the connection and select Make current beside a model. Edit corrects a model’s ID, and Remove takes it off the list.
- In the command palette (Cmd+Shift+P), run Switch AI model and choose a connection and model. That pair also becomes the default.
After Test connection, a model the server didn’t list is marked Not in the list this connection reported — check the ID. The provider will most likely reject it.
Cost and size limits
Section titled “Cost and size limits”When you send a request to an OpenRouter model, FlowBoard first works out what it is expected to cost from OpenRouter’s prices. If that’s more than your limit, FlowBoard asks before anything is sent: select Send or Don’t send. AI Assist then shows the estimate, with the most the request could cost if the model wrote as much as it’s allowed to.
Set the limit per connection under Advanced model limits. The default is $0.50, and 0 asks every time. Other providers report no prices, so FlowBoard can’t estimate their cost and doesn’t ask.
Before each request, FlowBoard asks LM Studio, OpenRouter and Anthropic how much text the model accepts, and fits the request to it. A model with a large context gets all the context you chose. For a smaller one, FlowBoard shortens or leaves out optional context, and AI Assist says which. For other servers, FlowBoard uses the Fallback context budget (tokens) under Advanced model limits.
- Draft, rewrite and ask shows how to use AI Assist on your cards.
- Saved prompts and story style keeps instructions you use often.