Tavus launches Dom, a computer-controlling AI assistant with a voice and avatar interface inspired by Apple's 1987 Knowledge Navigator

Everything in the demo was real and working: a single session, no mockups. I considered posting the raw footage from every angle just to prove it isn't faked. I'm printing the @ycombinator bottle adapter and will report back on its fit. SpotiPod is successfully syncing my Spotify playlists with my iPod Classic.

I’ve been using Dom for a week now and can’t express how much fun and intuitive he has been to use. I’ve completed projects I didn’t think I’d start because there was such low cognitive load.

It’s the small things though that I didn’t expect, and are what make him really feel like the future interface for computing. Opening Spotify myself isn’t hard. Writing an email isn’t hard. Even paying a parking ticket is emotionally painful, but not hard. Even working with Claude code to build a new app is something I could do myself.

But Dom makes the small things effortless, and the hard things seem easy. That’s the real magic.

The GUI wasn’t better because it could do more, it was because it could do the same things more easily. Dom, and the human computing interface he represents, feels like the next evolution of that.

Knowledge Navigator was amongst the sci-fi that promised a future beyond basic point-and-click interfaces, and certainly beyond the chatbots of today.

A future where using a machine feels more like talking to a friend or a coworker. You’d build trust and connection with it, and in turn it would get to know you, your preferences and proactively manage your life. Everyone would have their own Alfred Pennyworth.

At Tavus, we’ve been obsessed with working on this future of human computing, where machines learn our language instead of us learning theirs.

Dom is an early preview of this concept with a release planned for later this year (more on why below).

If you’re interested in getting early access, or just want to learn more about Tavus, feel free to email or DM me.

It's important to note that Dom is an early preview and not ready for general use. He still has plenty of rough edges, some of which you may notice in the video. His first cradle design for the iPod wasn't printable due to a floating cantilever. Once in a while he'll take over my computer trying to complete a task and won't let go no matter how much I beg him to stop. After a long enough conversation, his context can become poisoned and he'll lose the ability to use tools altogether.‍ Then, there is still more work to be done on compute efficiency and figuring out the right interface balance.

Some of these issues are engineering and harness problems. Others will simply improve with each new generation of models. We’re working on getting it all fixed and releasing Dom towards the end of the year.

One of those pieces is real-time intelligence.

A Knowledge Navigator-style assistant needs to reason, write code, generate files, use tools, and respond conversationally in real time. Until recently, there was also a brutal tradeoff between model intelligence and speed. Generally, models haven’t gotten faster as they’ve gotten more intelligent, they’ve gotten larger and slower.

This has been at odds with human computing because conversations are unforgiving. The amount of context and capability required for assistants like Dom meant that the only models intelligent enough were often models like Opus, whose TTFT made them impossibly impractical for a conversational interface.

To solve the real-time intelligence problem we teamed up with @cerebras. With Kimi K2.6 running on their inference engine, we’re getting frontier-model intelligence at real-time speeds: 1,000 tokens/s on a 1T parameter model.

In the demo, Dom wrote scripts, generated .stl files, and completed tasks before I had even finished my sentence. It’s an incredible unlock with our models and harness.

All of this together is the Human computing stack we’ve been building at @tavus : the face, the perception, the speech, the memory, the personality, the tools, and the intelligence behind it.

And you too can build on it to create your version of the future.