JevPilot
JevPilot (github.
Overview
JevPilot (github.com/standardagents/jevpilot) is Justin Schroeder's browser driving game inspired by Tesla FSD visuals. Geometry, collision checks, and a hard safety brake stay in TypeScript; Jev reads compact candidate steering and speed paths, not camera frames, and answers up to about four times per second near traffic. Play at jevpilot.standardagents.ai with server-metered credit, or clone the MIT repo.
Problem: Driving agents that read pixels burn budget and latency, while toy sims ignore safety when models get creative near traffic.
Built for: Game and robotics curious developers who want a Three.js sandbox where Jev steers from symbolic path tables.
First indexed on Jev Directory: 2026-09-16
Creator and team
- Name
- Justin Schroeder
- Organization
- Standard Agents
- Handle
- @jpschroeder
How Jev is used
- Role in the product flow
- High-frequency path Choice among filtered steering and speed candidates
- Primitives
- ChoiceScore
- State in
- Symbolic road boundaries, nearby traffic summaries, signal state, destination bearing, and tables of eligible maneuver candidates.
- Decision out
- Selected path index consumed by the sim loop; single-candidate cases resolve locally without an API call.
- Simulator samples steering and speed combinations each tick
- Code filters candidates that leave the road or collide
- Jev Choice picks among remaining paths (up to ~4 Hz near traffic)
- Safety brake overrides imminent collisions regardless of model output
Justin rebuilt the vibe of FSD as a teaching toy: the renderer is Three.js eye candy, but the agent loop is classic robotics factoring. Candidate paths are rows in a table; illegal rows never reach Jev; obvious singletons short-circuit locally. That is why the sim can breathe four decisions a second without sending pixels to a vision model. Press J to toggle autopilot in the hosted build. Credit metering on standardagents.ai keeps keys off the client, which matters when you hand the link to a classroom. The X clip is embedded here via remote twimg; README and build page carry the longer architecture narrative.
Sourced performance claims
- Author materials cite up to about four decisions per second near traffic and about 1.5 Hz on clear roads.Source: madewithjev.com/builds/jevpilot and x.com/jpschroeder/status/2100347770867458384
- Hosted demo meters about $0.25 of Jev credit server-side so API keys never reach the browser.Source: madewithjev.com/builds/jevpilot
- Public GitHub repo standardagents/jevpilot had about one hundred seventy nine stars when this listing was drafted.Source: GitHub star count September 2026
Features and stack
Features
- Three.js driving sim with autopilot toggle
- Symbolic path tables instead of pixel inputs
- Code-level safety brake on imminent collisions
- Hosted demo with server-side credit metering
Stack
- TypeScript
- Three.js
- TypeSafe Jev
- Standard Agents auth for hosted demo
Pricing: Open source repo; hosted demo includes about $0.25 metered Jev credit per build page notes.
Demo
Showcase clip with the same lightbox player as the homepage. Click to play.
JevPilot FSD-style driving sim
No pixel soup: tables of legal steering arcs, Jev picks a lane like a calm copilot, and the safety brake still owns your dignity near traffic.
Links
FAQ
- Does Jev see the canvas pixels?
- No. Public materials describe compact tables of candidate paths and traffic summaries, not screenshots.
- Can the model override the safety brake?
- Imminent collision handling is coded separately from Jev outputs per the build write-up.
- Is this a Tesla product?
- No. It is an open source homage and teaching sim from Standard Agents, unaffiliated with Tesla.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
- System One model
The model family behind Jev: parallel typed questions, one forward pass, probabilities you can threshold.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.