Rubidu is now $7,500.
A private AI voice assistant for the Mac. Speech, the model and the voice all run on the device, and it works with the Wi-Fi off. 5,391 tests.
App, source code, brand and accounts, through escrow. DM me or [email protected]
https://t.co/uKlVPeDFhB
I said I was building my own Jarvis for Mac and I'd need help 😭
Now it's real. A month-long journey later (I only started posting here late), it runs 100% on my Mac, and... I'm selling it 🫡
Why? I want to build my own PC, and I don't want my parents paying for it.
I built a private AI that lives in my Mac's menu bar.
Then Apple shipped Siri AI. Same idea, built into macOS.
So Rubidu changes job: it's the harness around the model. Ears, rules, safety, hands, voice. Offline, no account, 5,604 tests. The brain is swappable.
Just a ridiculous idea, in this era of expensive hardware(thanks to AI), how many of us are actually prepared for gta 6😭
The website is itself made by AI💀
https://t.co/DRlViYQYsM
#GTA6#gta#afford#FYP
Just a ridiculous idea, in this era of expensive hardware(thanks to AI), how many of us are actually prepared for gta 6😭
The website is itself made by AI💀
https://t.co/DRlViYQYsM
#GTA6#gta#afford#FYP
I'm testing an idea: a Mac app that reads any PDF to you in a natural voice, fully offline, one price, no subscription. Real sample on the page, no studio tricks. If 50 people want it this week, I'll build it.
https://t.co/M8LwGqEquv
The moment Rubidu was ready, Apple finally made Siri smart.
I could have stopped there. Instead I turned Rubidu into a smart harness: the ears, rules, safety and voice any AI needs, fully offline on your Mac.
Same passion. New job. #buildinpublic
https://t.co/Bcr8ST4xhz
The moment Rubidu was ready, Apple finally made Siri smart.
I could have stopped there. Instead I turned Rubidu into a smart harness: the ears, rules, safety and voice any AI needs, fully offline on your Mac.
Same passion. New job. #buildinpublic
https://t.co/Bcr8ST4xhz
I built a private AI that lives in my Mac's menu bar.
Then Apple shipped Siri AI. Same idea, built into macOS.
So Rubidu changes job: it's the harness around the model. Ears, rules, safety, hands, voice. Offline, no account, 5,604 tests. The brain is swappable.
I built a private AI that lives in my Mac's menu bar.
Then Apple shipped Siri AI. Same idea, built into macOS.
So Rubidu changes job: it's the harness around the model. Ears, rules, safety, hands, voice. Offline, no account, 5,604 tests. The brain is swappable.
I'm testing an idea: a Mac app that reads any PDF to you in a natural voice, fully offline, one price, no subscription. Real sample on the page, no studio tricks. If 50 people want it this week, I'll build it.
https://t.co/M8LwGqEquv
@MichaelGannotti push-to-talk on the same keys is the right default. talking to agents is great until your accent meets a model name 😅 local dictation got so good this year
@zexigh does the lead hold at the small end too? I run Qwen3-1.7B 4-bit on MLX for a voice assistant, and time to first token matters more to me than decode speed
@deeps4you "why is my monitor stuck at 30 Hz, in one sentence" is the feature I didn't know I needed. plain-word answers to macOS weirdness beat another settings panel every time
@jpaguiarmatos a calm nudge instead of blocking is the right call. I read the screen on-device with Vision in my Mac app, and for me the tricky part was deciding when NOT to say anything. how often does it check?
@randomisdope a petal dial on modifier keys is such a fun interaction. which Apple framework ended up doing the most heavy lifting across the 90 tools?
@mattt keeping a local copy and refreshing it over USB is such a nice touch. menu bar app + local index is the pattern I keep coming back to as well. how big does the cache get after a semester of notes?
Rubidu is now listed for sale ($7,500):
IndieMaker: https://t.co/gTXytRmM42
SideProjectors: https://t.co/tPAmeaMDXn
A private AI voice assistant for the Mac, fully on-device. App, source code and brand. Details: https://t.co/MCCZyo5E7C
@coletoncodes Mine is one repo, and I'm trying to sell it before it joins the graveyard 😅 Building was the easy part, getting anyone to look at it is the real job.
@hojoHQ@iam_chonchol I do both: keep the model loaded while someone's actively talking, then unload after a quiet stretch. On my Mac that's about 0.45 GB idle vs 1.7 GB ready. People notice a cold start much more than RAM.
@ScriptByAI Streaming sentence by sentence is the biggest perceived-speed win in voice. It's how my local Mac assistant gets its first spoken words out in about a quarter of a second.
@jmurillocode I measured on my own voice: whisper base.en got 87% of commands right, small.en-q5_1 93%, large-v3-turbo 97% (slower, more RAM). On macOS 26, Apple's on-device recogniser is free and very good too, worth trying first.