The panel appears over the app you are in, focused and ready. Nothing moves behind it. Record any other combination you like, in the first run window or in Settings.
An AI panel, one keystroke away.
It opens over whatever you are doing, answers, and gets out of the way. You bring your own API key, so you pay the model provider directly: no subscription, no markup, no account.
Then €6.99, once. macOS 26 or later · Apple silicon · signed and notarized by Apple
The panel over the app you were in. Answers stream as they arrive.
Three keys, and you are back to work.
No window to switch to, no tab to find, nothing left open behind you.
The answer streams in as markdown: headings, lists, code with a copy button. Press esc to stop it mid sentence, or ask a follow up.
The answer goes straight back into the app you came from, at the cursor you left. Or ⌘⇧C to copy it and leave.
Your models, your money.
Lume never sees a token. Every question goes from your Mac straight to the provider you chose.
No middleman, by construction
Most AI utilities resell you someone else's tokens at a monthly price, whether you asked anything that month or not. You paste an API key once, it goes into the macOS Keychain, and the bill is between you and the provider: itemised, and it stops when you stop asking.
One key, four hundred models
An OpenRouter key reaches Claude, GPT, Gemini, Grok, Llama, DeepSeek and Qwen. Or point Lume directly at Anthropic, OpenAI, or Ollama on your own machine, where a question costs nothing and never leaves the laptop.
Type / for the instructions you repeat. Each one can pin its own model: something fast to fix a sentence, something that thinks to explain a problem. Lume shows every model's price per million tokens, and how quickly it answers, while you pick.
Everything, for seven euros.
There is no other tier, and the trial is the whole app.
Quick prompts
Save the instructions you type again and again, fire one with /, then just paste the text. It stays attached to the follow ups until you take it off.
A model per prompt
Correcting a sentence and explaining a bug do not want the same model. Each saved prompt can pin its own, and the panel shows which one is about to run.
Web search, when it earns it
Switch search on for one question or for a saved prompt. Sources come back as small numbered pills, in place, and they are clickable.
The real model catalogue
Read live from the provider, not a list somebody typed once. Tick the handful you want in ⌘K and search the rest when you need them.
The key stays in the Keychain
Never in a config file, never on a server of ours, because there is no server of ours. There is no account to make, and what you type is never sent anywhere except to the provider you picked.
Works with no network at all
Point it at a local Ollama model and the whole thing runs on your Mac: no key, no bill, nothing leaving the machine.
Native, and light
SwiftUI and AppKit, no Electron, no bundled browser. Two megabytes, opens instantly, and does nothing at all while you are not using it.
Updates arrive on their own
Lume looks for a new version once a day and installs it when you quit. Every update is signed, and the checking has its own switch in Settings.
Signed and notarized
Developer ID signature with the notarization ticket stapled, so it opens like any other Mac app. The Accessibility permission is optional and only for pasting back.
Bought once, the way software used to be.
once, VAT included where it applies
15 questions, no card, no account. Buy a licence when you want to keep going.
- The trial is the whole app, not a cut-down one.
- Every feature. There is no other tier.
- Updates included.
- Use it on the Macs you own.
- No account, and nothing you type is ever logged.
- Model costs go to your provider, never to us.
The licence key arrives by email; you paste it into Settings once. It needs macOS 26 (Tahoe) or later on Apple silicon, and an API key from a provider of your choice.