whoa this actually worked! Jev lets me control my browser in real time with my voice now > i talk > transcript sent to Jev > jev returns probabilities in ~300ms > browser clicks costs: $0.0002 per decision i'm stunned how fast this is. when i asked it to... "go back", it even finished the request before i finished my sentence 😂show more

Nice experminent. I was also thinking about computer use. Would be great if it had multimodal inputs. Just a matter of time until it has, I guess. Currently experimenting with Jev as a classifier for a LLM autorouter. Also very promising here.

nice. so many interesting things to do, not enough time 😄

💯

oh, let me DM you quickly

I'll be doing a live Jev build session in my community for founders next week Join here:

Can you explain the setup? Or gh repo

@dhh - We need a Jev + Omarchy implementation, stat! This would be killer for physically disabled people who have trouble using their limbs to control computer input devices.

This really just makes me excited for when we get ASICs

Congrats. It even beat you to go back.

yes 😂

so jev is just a super fast classifier of intent? dangit, why haven't I gotten an invite yet!

Computer use might be solved dudes. So excited for SSI's model...!!

GLiNER does the same thing as Jev while being local, free and open weights

How do you get it to understand you want it to go to " You didn't put that specific URL into its decision tree did you?

You can control it with your voice? Amazing, it's the Clapper. AI is nuts :)

How do I see your profile now for the first time. Damn.

🫶

The weak part of the demo is task difficulty, not the model being used wrong. Browser MCP (existing) already gives us click/navigate/snapshot. Your voice mostly does that and could be implemented with a control prompt. What’s new? You've only swapped the LLM turn for Jev’s 300ms structured choice over the current DOM + partial transcript. Great for Jev. WebMCP is a different bet: where sites expose tools so you don’t click the UI at all. Your clip doesn’t test that, and Wikipedia isn’t speaking WebMCP. I'd like to see both.

Nice one. Next step pairing it with Neuralink

This is insane I didn't know Jev is this good. It's gonna be implemented in literally everything.

What kind of STT are you using? Is it local?

Jev has a very limited set of actions in response. How can it handle something like 'go to ? And it's funny you said .com and it understood it was .org !

I'm just wondering how you are taking decision rather than Jev taking the decision. You are telling do this to that, what is Jev doing here ? Try video give a high level goal and Jev decides and navigate of it own, true agentic browser use.

will you be sharing this, looks pretty useful.

yes will do

Impressive demo!

Jev looks great for computer use! Loving these examples coming out!

I got access today and immediately had no idea what was going on.

Latency is impressive, considering it also includes STT – what are you using for that? Local model?

cool use case man! thanks for sharing, getting ideas for how we might use this - any other applications of this?

Instead of using your hand and clicking for free? I mean, okay, I can imagine proper, useful use cases, like helping people with disabilities.

@FarzaTV

does it work via API?

Wait I'm a bit confused, I thought this thing was a big probabilistic if statement. How does it recognize each actions, do you send the hotel page or something?

incredible

great usage. will it soon be time to put away the mouse and keyboard?

Setup? Need to test it with my Jev.