Framer AI AgentsDesign and publish professional sites with AI
Lumo
Your inbox, handled.
5 followers
Your inbox, handled.
5 followers
Lumo is a Windows desktop AI email chief of staff. It reads every email as it arrives, files the noise, and gives you a running brief of what's actually happening in your inbox. It drafts replies in your voice using knowledge from your real email history — and its signature rule: it never fabricates your opinion. If a reply needs a decision only you can make, it asks you instead of guessing. Auto-send is off by default, every send has a 5-minute undo, and replies thread properly.
One thing I'd love is a way to set per-contact voice profiles, so the draft tone shifts automatically when I'm writing to my boss versus a close client. Right now it sounds consistent across everyone, which doesn't match how I actually write. Would make the replies feel even more like me.
@berke746137 Love this comment, because you've described the exact design goal — and you've caught the honest gap in it.
Under the hood, Lumo already learns voice per relationship: it builds a pattern atlas from your sent mail that tracks how your formality, openers, and signoffs shift by tier (inner circle vs peers vs cold contacts), and when it drafts to a specific person it uses your actual past replies to that person as examples. Your reply to your boss and your reply to a close client shouldn't be built from the same voice at all.
The catch — and why it sounds uniform to you right now: all of that needs history. Early on, before it's seen enough of your sent mail with a given contact, drafts regress toward your average voice, which is exactly the flatness you're hearing. It gets more "you-per-person" the longer it runs.
But a manual per-contact voice override ("always formal with X, always casual with Y") is a genuinely good idea — it fixes the cold-start without waiting for the learning to catch up, and it's very shippable. Adding it to the pilot roadmap.
How long has it been running on your inbox? If you're past the first scan and boss-vs-client drafts still sound identical, I'd love to dig into your specific case — DM me.