Case wall
Jev AI Agents cases, page 3
107 public posts filed under Agents. English summaries on this page, with a link to each case.
Last updated
These are the public posts filed under the Agents category.
No posts match these filters.
@MadisonJRickert built a Jev classifier for Claude Code auto mode and says the permission checks it decides are 2x faster.
Anthropic released native mods for customizing Claude Code yesterday, so I did the obvious thing and built a Jev-based classifier for auto mode! 2x faster permi…
"@N01ennn shows a RAG loop where one Jev call scores 20 passages + answerable/injection gates, then a second Jev call grounds claims."
this is pure f*cking treasure A Stanford AI research group finally drew the perfect RAG system: retrieval, Jev and agents in one loop, and it fixes the 3 things…

@DominikTornow's stfu asks Jev on each prompt how long Claude's reply should be, then adds that limit to the turn.
Shut up Claude with @typesafeai. On each prompt, Jev decides how long the reply should be and adds a length limit to the turn. Get an answer, not an essay. http…
"@gippp69 routes 20,000 events/day with Jev so only 4% reach a writer: Jev $0.84, full loop $29.04."
Jev builders: the model bill is decided before the model is called /routing 20,000 events/day become: → 14,000 die as noise → 4,700 become logs → 800 earn a dra…
@browser_use shows Luna as the planner and Jev as the actor, playing poker and winning.
Luna (planner) -> Jev (actor) playing poker and winning
@orcdev says TanStack AI shipped subagents and that Jev picks which models or agents should run for a task, in parallel, in sequence, or mixed.
TanStack AI just shipped subagents. Jev picks which models / agents should run for the task, in parallel, in sequence, or mixed. Best Jev use case I've seen so…
@Dain0x sketches a workflow to test: a pipeline finds candidate signals, Jev picks which to investigate, Opus explores explanations, then stats and human review ask what evidence would change the conclusion. The animation is described as simulated.
A beautiful pattern can still be noise. Here’s a JEV × Opus 5.5 workflow I’d test: → A data pipeline identifies candidate signals. → JEV selects which candidate…

@fladdict built a harness that uses Jev to watch Codex and stop bad implementations or runaway behavior in a Codex process meant to run 24 hours. They say it seems to be working and an endurance test is in progress.
JevでCodexを見張って、24時間稼働のCodexのクソ実装や暴走をとめるハーネスを作ってみた。わりと上手く動いてるっぽい。耐久実験中。
@MrAhmadAwais says Jev launched on Command Code for free, usable in headless mode or via the provider API, and that they open-sourced a mod, cmd-mod-jev-nudge, so Jev can keep an agent running while more work remains.
Launching Jev on Command Code for free today. Jev x Command Code is super interesting. You can only use Jev in headless mode or Provider API. Also open sourced…
@codeaholicguy built a session compact for AI DevKit with Jev labeling each message so code keeps only what the next agent needs (1.72M → 5.9K tokens).
I built a session compact for AI DevKit, using Jev to label every message, and code keeps only what the next agent needs. Tested on a real session 1.72M → 5.9K…
@tamarajtran argues compaction should not be a summarization prompt. They use TypeSafe Jev to score every tool call and drop what is irrelevant, which they call instant compaction.
found the perfect use case for @typesafeai Jev: instant compaction in 2026, why is compaction still a summarization prompt? Jev can make it instant by scoring e…
Quoted @thread
@PaliwalMuskan19 built a skill-picker because keeping 30 or more agent skills meant remembering which one does what. They say Jev (@typesafeai) ranks those skills for a given task and tells them and their agents which skills are actually worth using. The repo is github.com/MuskanPaliwal/skill-picker.
built this skill-picker for myself because apparently having 30+ agent skills also means remembering which one does what 😭 jev (@typesafeai), being the god that…

@CartwrightApp built Cortex, a local MCP server: Claude plans, and a fast layer does the clicking. On the same Mac-app task they measured computer use at 23 seconds versus Cortex at 0.8 seconds, and one action from about a 5 second Claude round to 0.25 seconds with no model. They list 24 tools, including a browser, Mac apps, GitHub search, and a safety gate, say it was measured on their Mac, and tag @typesafeai and #JEV.
Built Cortex, a local MCP server: Claude plans, a fast layer does the clicking. Same Mac-app task: computer use 23 s → Cortex 0.8 s One action: ~5 s Claude roun…

@kunchenguid introduces compact-adviser, an agent plugin for Claude and Pi that uses Jev to judge whether a session is at a task boundary safe to compact. They built a private eval from 40 real sessions and say the classifier favors precision when the window is small and recall as it fills up.
almost every day i hear people ask "when should i /compact my session" there's no easy answer because it depends on how likely your future action will need deta…
@Saccc_c compares Codex’s built-in computer use with a “Jev Use” setup that pairs Codex and Jev. In a video of adding a Mac calendar event, they say the Jev version has almost no pauses and similar token use.
卧槽,居然真有比Codex内置computer use更快的操作电脑的方式 我尝试用Codex+Jev打造了一个加强版computer use,我称之为「Jev Use」。比内置的更快更丝滑,token消耗却差不多 下面是我用「添加Mac日历事件」做了一个对比视频,同样的内容,明显可以看到Jev版本整体过程几乎无任何停…
Quoted @Saccc_c
@gregpr07 built a small open-source browser agent with Browser Use and Jev. Finding flights took 7 seconds and cost $0.0039. The action space and DOM state change every step, with a small LLM fallback for typing. The video is at 1x speed.
Breaking: Browser Use + Jev = Ultrafast ⚡ Findings flights took 7s and cost only $0.0039 🤯 > new action space every step > DOM state space > small LLM…
Quoted @thread
@redp314 reviews pull requests with one TypeSafe call that returns 14 probability checks, including hardcoded secrets, SQL injection, auth, deleted tests, API breaks, migrations, debug leftovers, whether the description matches the diff, blast radius, and reviewer effort. Code turns that into BLOCK, security review, nits, or merge, and uncertain critical checks (0.35–0.65) escalate. They cite 6 real PRs at $0.00007 each.
got Jev to review my PRs. ~200x cheaper than Claude and it answers in half a second 6 real PRs in the video. $0.00007 each. 1,000 PRs = 7 cents vs ~$14.50 on Op…
@moritzkremb controls a browser by voice: speech is transcribed, Jev returns probabilities in about 300ms, and the browser clicks, at $0.0002 per decision. They say a “go back” finished before they finished the sentence.
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…
Quoted @thread
@o_kwasniewski is building tester.army, an open-source framework for end-to-end tests with agents, including web and mobile, and shows it with TypeSafe Jev. They say it will be available soon.
e2e + jev from @typesafeai ⚡ I'm building an open-source framework for running e2e tests with agents. supports web, mobile (and more!) available soon: https://t…
@milindlabs does computer use without screenshots or an LLM leaving the Mac: a local CoreML model segments buttons, on-device OCR reads labels, and Jev returns a probability over those elements and clicks in a loop until the goal is done. They report about 90ms per decision.
Okay so Jev can actually do computer use really well Without any screenshots, or LLMs and no Pixels leave my mac I dont even read the Dom elements A local CoreM…
Quoted @CompleteSkeptic
@instantricecook built voice-controlled computer use for a Mac with TypeSafe Jev. They say they can dictate “open the Notes app and create…” and the app opens before the sentence finishes.
I built a voice controlled computer-use for my mac using @typesafeai's Jev and it's INSANE how fast it is! I can dictate "open the notes app and create..." and…
@camsoft2000 shows TypeSafe Jev plus AXe controlling the iOS simulator, which they call ultra fast and a fraction of the cost of an LLM.
Now we're cooking. Got @typesafeai's Jev + AXe controlling the iOS simulator now ultra fast at a fraction of the cost of using an LLM. This is game changing! ht…
Jev chooses an action from the accessibility tree, then Stagehand executes it.
we built blazing fast computer/browser use with Jev + @Stagehanddev. this task cost $0.001 and executed at near instant speed (in a remote browser btw) the loop…
@savboj posts that Jev plus computer use is 100x faster than any LLM, and that if you blink you might miss it. The text does not describe the task.
If you blink you might miss it Jev + Computer use = 100x faster than any LLM @typesafeai https://t.co/JG09Wx7UCo
Quoted @CompleteSkeptic