Brother, creator, award-winning singer songwriter. Leading marketing and content at @desertantlabs.
North of Ireland
Joined April 2008
- Tweets10.3K
- Following3.1K
- Followers4.7K
- Likes17.6K
Pinned Tweet
Very excited to have joined this team.
Today we're launching Desert Ant Labs: a European frontier AI lab building on-device intelligence.
18 models across audio, vision, and text. SDKs for Swift, Kotlin, and JavaScript.
No tokens. No logins. Nothing leaves the device.
desertant.com
Daniel Duke retweeted
AI emoji suggestions you don’t have to wait for.
Emo suggests emoji based on what you type, on every keystroke, on device.
On an iPhone 17 Pro, Emo takes 1.7 ms per suggestion, fast enough to keep up with you.
22 languages. 5MB. Nothing leaves the phone.
desertant.com/models/emo
Daniel Duke retweeted
Typing to your agent is so 2025. Here are three dictation apps built with Voz, so you can talk instead.
∗ Takt puts push-to-talk on Windows: hold Right Alt, talk, and the text lands in whatever app you’re in. Clear cleans up the mic audio first.
∗ Whisper Anywhere does voice typing from the Windows tray, with Voz as the default, and never writes a recording to disk.
∗ OpenFlow brings it to Android as a floating bubble that shows up with the keyboard, with Ear detecting the language.
Every word is transcribed on device. Links in the thread.
Daniel Duke retweeted
Half as many false flags as NudeNet.
Moderator is a 9.7MB NSFW detection model that can run on a phone before a photo reaches your servers.
In our tests, it wrongly flagged 6.3% of safe images, against NudeNet’s 12.9%, and caught 87.8% of NSFW images.
desertant.com/models/moderat…
Astra vs Opus 5.5 is such a stark contrast. Astra feels like directing an intern, Opus 5.5 feels like collaborating with a peer.
Astra can't even do consistent work in an existing codebase and it just devours usage limits.
Daniel Duke retweeted
A cloud API call doesn’t need your contacts’ names.
Cue, @AgrawalHridaya’s smart clipboard for Mac, strips them out before anything leaves your machine. Redact runs on the device and swaps names, phone numbers and addresses for placeholders, so the cloud model sees [GIVEN_NAME_1], not Maya.
The paste still uses your real text, and if Redact can’t run, Cue does a plain paste instead of sending anything unredacted.
Very cool.
github.com/hridaya423/cue
Daniel Duke retweeted
Built with @pidotdev durable, the agent loop in Pelican runs entirely on your iPhone, using any provider you want, remote or local. Under the hood a lot of harness engineering has gone into it to make sure it's super fast, token-efficient, and secure.
Things like smart permission dialogs only after the agent has ran a tool that could possibly be a prompt injection surface, a built-in web search tool that gathers search results on your phone and parses them with the same html-to-markdown parser that powers @obsdmd web clipper, routines as durable runs that check in on things in the background on your phone, and even a fully durable chat sync across devices using iCloud.
Exactly four people said they wanted this so obviously I built a little landing page where you can download it for free: littlevictori.es/hermes/
Daniel Duke retweeted
hey sam, your mclaren is only used 5% of the time. this is a terrible allocation of scarce resources. have you considered taking the bus to work
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
I'm constantly fighting my non-airpod headphones when I want to switch devices so I built a little menu bar app that lets me steal them from my other Mac with one click.
And when a Mac locks or sleeps, it lets them go automatically so I can pick them up on my phone instead.
How it works: Hermes runs in the menu bar on each Mac. You link them once with a short code, and after that they find each other on your local network.
Daniel Duke retweeted
Your AI agent should live in your pocket, not on a Mac mini.
Pelican runs on your iPhone, works with Siri AI, and does real things in Calendar, Reminders, Mail, Photos, and more. Use any AI you want. No account.
Pre-order now. Out November 9.
Daniel Duke retweeted
A quarter of a second. That's how long four of our on-device models took to turn a 12-second voice memo into a named, sorted note.
Voz transcribed the audio, Title named it, Emo picked an icon and Gist filed it under a topic.
Chain them together in your app and you have a voice notes feature where the recording never leaves the device.
Daniel Duke retweeted
T-minus 6.4 seconds.
That's how long our speaker diarization model, Who, took to label every speaker when we tested it on a 57-minute NASA podcast.
It ran at 530× real time, on a MacBook, with nothing leaving the device.
Daniel Duke retweeted
Andris at @BetterStackHQ put our new Who and Eye models to the test.
Who labelled every speaker in their podcast in 11.9 seconds, 393 times faster than real time, on an M2 Max.
Eye sorted 105 photos in 16 seconds, right in the browser.
All audio and photos stayed on the device.
youtube.com/watch?v=7oWM2UEH…
Daniel Duke retweeted
The iPhone Duo is out today, and @detailapp's Auto Edit runs on it entirely on device.
Detail edits a 10-minute podcast recording in 8.4 seconds: a full edit plus a dozen shorts, each with a title and its own transcript.
Six of our models do the work: Ear detects the language, Voz transcribes, Clear cleans up the audio, Uhm cuts the filler words, Clips picks the best moments and Title names the clips.
desertant.com/models
Daniel Duke retweeted
Let your camera or gallery app pick the best photo from a burst. Eye finds the frames where someone’s eyes are closed and keeps the best of the rest, without the photos leaving the phone.
Eye is in version 4.0 of our SDK for Swift, Kotlin and JavaScript: desertant.com/models/eye
Daniel Duke retweeted
You asked for speaker diarization, so we built it.
Who split this 43-minute podcast by speaker in 4.9s on an M5 chip, including the moments two people talk at the same time. The recording never left the machine.
Who is in our Swift, Kotlin and JavaScript SDKs, free up to 100k monthly active devices: desertant.com/models/who