lepach

formerly @google @fitbit @pebble

[lepach @_lepach]

@NoemiTitarenco That is my take as well, the hard part of what you've made is having the agent be proactively watching for changes but being local.

[lepach @_lepach]

@jessfraz It's likely because humans provide signals they capture to improve their systems and decide what to feature.

I'd guess that Amazon doesn't want to become like an API because then they lose a ton of soft control over the market and the advertising they build into it.

[lepach @_lepach]

@JoshKuechly @nvidia @Meta @OpenAI The novel thing is zero shot classification with a decent API shape that is fast and cheap. I don't expect Jev alone to dominate but it unlocked people's imaginations.

[lepach @_lepach]

@Cryptonaut14 @thekitze The full list of skills available, each is considered separately based on the current context.

[Matt DesLauriers @mattdesl]

does Jev understand colour?

incredible model, very cheap and fast. definitely feels like a leap forward for creating new UX/UI paradigms on top of modern language models.

[lepach @_lepach]

@_Neo_3 @nikhilmudholkar That's basically what I was saying but it didn't come out right. Use it as the first layer and when certainty drops below a set threshold fallback to Gemini.

[Kun Chen @kunchenguid]

Jev gave me $5 of API credit when i signed up

when i die, i will pass what's left of it to my son

[lepach @_lepach]

They may or may not be aiming for it, but it is their end game: EverythingStack

Meaning you use your BigLab Device which has your fully customized BigLab OS experience. Not just going for Apple ecosystem, but to also be the marketplace for unique experiences that people share and maybe can sell all on BigLab OS!

[lepach @_lepach]

@pvncher I think it is interesting, but with some creative macros you could likely create something similar for Rust.

[Michael @michaeltefula]

You can use Jev to assemble interfaces on the fly.

Here’s an example I hacked together with Astra.

In this demo, Jev turns natural language requests into UI plans in roughly 1 second or less.

It’s 6x faster than Gemini 3.8 Flash and GPT 5.6 Luna. Cheaper, too.

[lepach @_lepach]

@kunchenguid That's what they do though, see what people make and then sherlock it. Most people using their stuff don't really know to look elsewhere or care to be vendor-agnostic.

That said I am with you about avoiding lock in. I don't even use other people's skills, remix into my own.

[lepach @_lepach]

@maria_rcks It is more a solution to cheaply identify calls that can easily be approved without an LLM. The trickier calls like dense opaque code mode tool calls will be harder for it to fully understand. Especially if it refers to opaque values like variables not directly defined.

[lepach @_lepach]

@VictorTaelin Hit this issue due to a submit failing and being too long.

Would be interesting to have an OpenAI level swarm attempting to break some laws with Bend2

[Armin Ronacher ⇌ @mitsuhiko]

This morning’s AI Twitter is “Drop everything people. New catnip dropped.”

[lepach @_lepach]

@dexhorthy I never use other skills directly, always have them evaluated and rewritten for my setup. I have a version of show me which points to things I already was doing similarly but it still is a great simple skill so thanks for sharing!

[lepach @_lepach]

@alonzuman Count me in!

*On both 😂

[lepach @_lepach]

@mitsuhiko Given the cache discounts I wonder if it would make sense to have an idle cache warmer that makes a small request that shouldn't output much but keeping the cache prefix warm. Just ignore the response and only fire when there's an active tool call waiting to finish.