Slipway's AI features use your own provider account. You choose the provider and model in Settings ▸ AI, and every request goes directly from your Mac to that provider. Nothing passes through a Slipway server, and Slipway doesn't add any charge on top of your provider's.
What you need#
- An API key from one of the supported providers, or Ollama running on your Mac.
- For a custom endpoint: its base URL and a key.
Supported providers#
| Provider | Default base URL | API key |
|---|---|---|
| Anthropic | https://api.anthropic.com/v1 |
Required (sk-ant-…) |
| OpenAI | https://api.openai.com/v1 |
Required (sk-…) |
| Google Gemini | https://generativelanguage.googleapis.com/v1beta/openai |
Required (AIza…) |
| xAI Grok | https://api.x.ai/v1 |
Required (xai-…) |
| Ollama (local) | http://localhost:11434/v1, editable |
Not needed |
| Custom (OpenAI-compatible) | You enter it | Required |
Anthropic is called through its Messages API. Every other provider is called through the OpenAI-compatible chat completions API, so Custom works with any service or local server that speaks that format.
Set up a provider#
- Open Slipway ▸ Settings… (
⌘,) and choose AI. - Pick a Provider.
- Paste your API key. For Ollama or Custom, fill in Base URL too; Ollama uses
http://localhost:11434/v1if you leave it empty. - Wait a moment. Shortly after you stop typing, Slipway asks the provider for its model list and shows, for example, 38 models from OpenAI.
- Click the chevron next to Model and pick one, or type any model ID into the field.
- Check that Status says Ready.
The key is saved in the macOS Keychain. The provider, model and base URL are saved in Slipway's settings.
The model list#
Slipway fetches the list from the provider's /models endpoint and hides models that can't chat, such as embedding, speech, transcription, image, video and moderation models. To fetch it again, click the circular arrow next to the model field ("Fetch the model list from …"). The list isn't required: you can type any model ID your provider accepts.
Switching provider clears the model, because model IDs belong to one provider.
Status messages#
| Status | Meaning |
|---|---|
| Ready | AI features can run |
| Needs an API key | Enter the provider's key |
| Needs a key and base URL | Custom needs both |
| Pick a model | Choose or type a model ID |
Which features use AI#
| Feature | Where | Button | Plan |
|---|---|---|---|
| Translate a review into English | Reviews | Translate with AI | Free |
| Draft a reply to a review | Reviews | Draft with AI | Free |
| Turn reviews into features, bugs and improvements | Reviews ▸ Insights | Analyze with AI | Free |
| Suggest search terms | Optimize | Suggest terms with AI | Free |
| Draft a reply to App Review after a rejection | Submit | Draft with AI | Free |
| New headline, subhead and background for one screen | Screenshots inspector | Regenerate This Screen | Free |
| Translate a listing into every other locale | Metadata | Localize | Pro |
| Rewrite subtitle, keywords and promotional text | Optimize | Rewrite with AI | Pro |
| Regenerate every screen in the set from a brief | Screenshots inspector | Regenerate All … Screens | Pro |
| Write captions for every locale, then render every size | Screenshots inspector | Generate All Sizes & Locales | Pro |
| Carry out an instruction in App Store Connect | Any app page | Ask AI field | Pro |
Pro — The five Pro features need Slipway Pro as well as a provider key. During the beta the paywall unlocks locally. See Pricing and Pro.
Most AI output is a draft. Review replies, App Review replies and Rewrite with AI copy appear in Slipway for you to edit, and nothing goes to Apple until you send or push it yourself. Localize is the exception: it translates the current locale into every other locale and pushes each one to App Store Connect as it goes. See Metadata. Ask AI also acts directly, and asks you to confirm destructive steps. See Ask AI.
What is sent to the provider#
Each request includes a short brief of your listing from the primary locale: the app name, subtitle, keywords, the start of the "What's New" text and the start of the description. Each feature then adds what it works on, such as the review being answered, the reviews being analysed, the locale being translated, or the screen's current headline and background.
Your App Store Connect key is never sent to an AI provider. See Privacy and security.
Tips#
- Ollama must be running, and the model must be pulled first, for example
ollama pull llama3.1. Nothing leaves your Mac. - Custom always asks for a key. If your local server doesn't check keys, enter any placeholder text.
- Smaller, faster models work well for review replies and translation. Use a stronger model for Localize and Analyze with AI, which handle longer text.
Troubleshooting#
- "Add your AI provider key in Settings first." The provider, key or model isn't set. Slipway opens Settings for you; fix whatever Status reports.
- "Couldn't list models: …" The key or base URL is wrong, or the provider is unreachable. The message after the colon comes from the provider. You can still type a model ID.
- "No chat models returned." The provider answered, but every model was filtered out. Type the model ID yourself.
- "… ran out of output tokens before answering (reasoning models need a larger budget)." Reasoning models can spend their whole output budget thinking. Pick a non-reasoning model for that task.
- "AI did not return JSON." Most features ask for a structured answer. Some small or local models don't follow that reliably; try again or choose a more capable model.
- Requests take a long time. Slipway waits up to five minutes for a reply, because reasoning models can be slow on long inputs.