THE JEV GUIDE · CURATED EDITION
Launch news, explanations, demos, and community projects around TypeSafe Jev, organized by topic. Every entry links back to its X source.
Curated selection · Not exhaustive · Updated manually
01 / EXPLORE
0 means awaiting a verified post
The author asks whether anyone uses Jev-style models for dynamic control systems and wants to test if such a model can balance an inverted pendulum.
At PromptQL's "Rebuild With Jev" hackathon, the team spent 90 minutes exploring whether TypeSafe AI's Jev can supervise a robot in real time, and won.
A post showcasing a trash robot controlled by Jev, described as a Jev version of a robot duck, with a verification link. Link
Automata Room tests Jev as the decision maker for a Unitree G1 humanoid, tasked with carrying 4 crates one at a time to a storage rack then returning to the entrance, with fewer shorter trips scoring better.
Jamieson Becker notes that the #Jev model is small and fast enough to run on a battlefield robot, even play DOOM, raising thoughts about fully autonomous robots in real life.
Introduces the Jev Drone Simulator created at the Jev Hackathon, using Jev for decision-making and control input determination, including automatic mode switching between standby, takeoff, normal flight, obstacle avoidance, and shoot-down mode.
The author suggests wanting to try having Jev make driving decisions given the current trend.
A user notes that Jev already runs 15 autonomous drones at once, suggesting driving a single car would be easy for it.
The author shares a bio-robot hybrid that uses Jev as the decision layer between a simulated nervous system and the robot's physical movement.
Robert connected Jev to a bio-inspired robot running on a simulated nervous system: neural activity feeds into Jev, Jev makes a rapid decision, and that decision becomes physical movement.
The author gave Jev access to an unreleased world model and a local segmentation model to test autonomous driving despite being unable to see, with results resembling Crazy Taxi but surprisingly functional.
A developer attempts arm control using only JEV, without pre-trained data, policy, or LLM, using two layers to detect motor and object states, and notes JEV currently lacks image input support.