jev has changed computer-use forever. i built a computer use tool with jev that costs basically nothing to use, and has sub-second latency. it can even decide what to type, and doesn't need an LLM. open source. link below. no LLM required.

repo: also, if you're building computer-use agents that need applications without the overhead of an entire computer, i'd love to talk - @tryisle

Wait I thought Jev could not generate text, so how does it type?

Nice! Consider adding it to

Great proof that computer use can be reduced to fast, bounded click/type decisions instead of a generative loop. We used that split for social research: Jev picks typed actions, socai runs them in Chrome, and evidence becomes a cited report:

here are some opensource use cases, love if you can submit a pr here as wlel

Congrats man, but at this point every second person who has access to a coding agent did it

add this on

The real breakthrough is making computer use an engineered primitive instead of an expensive LLM loop.

wait, wdym "doesn't need an llm"? @grok

Hey saw your repo, wanted to talk about the pros/cons of using screenshots/labeling ui elements vs using internal accessibility stuff the apps expose. Would love to talk more

Could you make it test mobile games/apps somehow? So it tests everything, clicks everywhere, tries to break it and notes all the bugs?

How good is it at tasks though?

The no-LLM part is the unlock

how does it decide what to type without using an LLM?

Im gonna install this to help with youtube and opusclip automation and see if it works

wait so this can actually be your personal assistant like actually

sub-second: relating to a time interval of less than one second

sub-second latency on computer use is the part that matters. nobody talks about it but it's why most of these demos feel broken in practice

nice one 🔥🔥🔥

Nice! How are you building the choices for what to type? My Codex version supplies the text and lets Jev choose the action—curious about your approach.

Damn JEV is like the upgrade from cydia repos x 10000

can't see yet only receives DOM

Wait, no LLM at all?

latency was always the bigger blocker for computer use, more than cost. a screenshot → model → action round trip on every step adds up fast on a 20 step task. drop the model call and most of that goes away. then the hard part is grounding: finding the right element after the UI

the important distinction is not whether a model is in the loop. it is whether the system can keep the action narrow and check the result before it becomes consequential.the important distinction is not whether a model is in the loop. it is whether the system can keep the action narrow and check the result before it becomes consequential.

No LLM in the loop. The keys already knew the sentence.

Can it fill out forms for me? Can it send text through whatsapp?

The app keeps quitting the app I want to interact with

不靠大模型也能把下一步选出来 这才是真正的 agent 基础设施 延迟和成本一起下来 普通人也能随手接进工作流

how yall make all dis things

Thats cool and all but wouldnt navigating via dom insta flag you as a bot? Especially since it is so incredibly fast

Would you have any extra invite?

This is so interesting

Ew why did you make him do such an abhorrent act to your machine