Jev posts

Describe a mood and Jev pulls matching typefaces for inspiration.

Whisper transcribes locally and Jev maps looser phrasing to the right keybinding. Say "computer" or hold F10. Open source.

A React site search that re-ranks results with Jev as you type. Playground and repo in the thread.

Every tool call is judged before it touches your machine: rules block the known-bad in ~40 ms, Jev decides the rest (allow, ask or block).

Layouts spoken into existence: a local transcription model, shadcn components and Jev deciding what goes where.

Wired into MiniMax H3: Jev scores each layer and picks a 1–10% attention sparsity. On an RTX 4070, 6:07 dropped to 3:34.

Metix AI retrieves the jobs and Jev decides the fit for each resume–job pair, 20 requests in flight.

Jev is cheap and fast enough to live inside consumer products, games included. Here it plays Codenames.

Static analysis picks the code, then Jev answers yes/no checks with a probability, like PII in logs or getters that change state.

A smarter copy and paste, built on Jev. The author thinks every computer interaction will get rewritten like this.

Jev scrolled 723 ads as 30 buyer personalities: 21,690 stop-or-scroll decisions for 22 cents.

Jev classifies thousands of listings by architecture, renovation status or distance to freeways in under 20 seconds, for $0.18.

An honest experiment: the game needs ~70 ms, Jev averaged ~650 ms, so the game pauses for each move and the waits are cut out.

Jev decides UP or DOWN on Polymarket's 5-minute Bitcoin market; the video shows a profitable call. Open source.

Stop defaulting to the biggest model at max effort: Jev picks the model and effort for each request in Codex and Claude Code.

OpenAI's Luna builds the list and Jev handles the fast decisions to fill a real Giant Food Stores cart, cheaper than a frontier model alone.

Two models steer their own dinosaurs on the same track: dodging, ducking and using shields, while Jev also throws the obstacles.

An open-source Chrome extension that replaces exact-match find with Jev, near real time.

Type a few words and Jev finds the Mark Rothko paintings that match.

A backlog of hoarded X bookmarks, run through Jev: light work for a model this fast and cheap.

Jev drives the player (keys and mouse) while a planner sends it instructions and writes new skills after each failure. Under $1 per run.

Talk and Jev reads the transcript and what you're wearing, picks from your closet and changes the outfit live. ~620 ms and $0.0011 per decision.

Jev filters content and selects the assets; DeepSeek and Higgsfield turn them into ad creatives.

Fourteen yes/no questions per post, like "does the hook open a loop?": $0.67 in total, about 680x cheaper than Opus per post.