AI setup.
anf's AI (summarize, ask, suggest names, auto-tag, organize, image search) is off by default. Turn it on and connect one model — Claude is the easiest and most capable.
Which one?
Three backends. Pick by where your data may go and what hardware you have.
| Provider | Where it runs | What you need |
|---|---|---|
| Claude — recommended | Cloud (Anthropic) · pay-per-use | API key + billing |
| Local LLM | Your Mac · free | Ollama or LM Studio running |
| Apple | Your Mac · free | macOS 26 + Apple Intelligence |
"aiProvider": "auto" tries Claude (if a key is set) → a local server (if one is up) → Apple, in that order. Pin one explicitly with "claude", "local", or "apple".
1. Open the settings file
In anf press ⌘, or Tools menu → AI Provider…. Edit the JSON, then switch back to anf to apply.
Claude (recommended)
A 1M-token context handles whole documents and folders. It's cloud — content is sent to Anthropic.
- Create a key at console.anthropic.com → API Keys. It looks like
sk-ant-api03-…(ask-ant-oat…OAuth token won't work). - Add a payment method / credits (Billing) so calls are allowed.
- Put this in the settings file:
"aiFeatures": true,
"aiProvider": "auto",
"aiApiKey": "sk-ant-api03-your-key"Leave the model empty for the default claude-opus-4-8 (deepest), or set "aiModel": "claude-sonnet-4-6" for faster, cheaper replies. Billed per token by Anthropic — a summary is a fraction of a cent; whole-folder questions cost more.
Local LLM (Ollama · LM Studio)
Stays on your Mac. Works without Apple Intelligence.
- Start the local server — Ollama runs automatically (
:11434); LM Studio: Developer → Start Server (:1234). - Settings file:
"aiFeatures": true,
"aiProvider": "local",
"aiEndpoint": "http://localhost:11434/v1",
"aiModel": "llama3.2"LM Studio is http://localhost:1234/v1. aiModel is the loaded model's name (LM Studio can leave it blank). Good picks: llama3.2, qwen2.5, or an 8B-class model — bigger is smarter but slower; plain (non-reasoning) models feel snappiest.
Apple on-device
No bundle, no network. Needs a Mac with macOS 26 + Apple Intelligence.
"aiFeatures": true,
"aiProvider": "apple"Turn on Apple Intelligence in System Settings; the Siri language must match.
Settings reference
Every AI key lives in the same settings file (⌘,).
| Key | Values | Default |
|---|---|---|
| aiFeatures | true / false — the master switch | false |
| aiProvider | auto · claude · local · apple | auto |
| aiApiKey | sk-ant-api03-… (Claude only) | — |
| aiModel | any model name for the chosen provider | claude-opus-4-8 |
| aiEndpoint | http://localhost:PORT/v1 (local only) | — |
Cloud (Claude) sends content to Anthropic. Local and Apple never leave your Mac.
Troubleshooting
- AI actions are greyed out →
aiFeaturesisn'ttrue, or you didn't switch back to anf to reload the settings file. - Claude: 401 / "invalid x-api-key" → you pasted an
sk-ant-oat…OAuth token. Use ansk-ant-api03…API key. - Claude: "credit balance too low" → add a payment method or credits under Billing.
- Local: "connection refused" → the server isn't running, or the port is wrong (Ollama
11434, LM Studio1234). CheckaiEndpoint. - Local: "model not found" →
aiModelmust match a pulled/loaded model (ollama list). - Apple: unavailable → needs macOS 26 with Apple Intelligence on, and Siri's language must match your content.
- Replies are slow → local reasoning models think a lot; pick a smaller plain model, or use Claude.
Still stuck? Open a GitHub Issue.
How to use it
- ⌘K →
/→ ask the current folder right away - Right-click a file/folder → Summarize · Ask · Suggest Name · Tag
- Right-click empty space / the Tools menu → folder summary, ask, organize
Image search, OCR, and tidy/organize-by-kind work without Apple Intelligence. Problems? Open a GitHub Issue.